Added AI disclaimer to README.

Export used to suggest "<name>_synced<ext>", losing the offset value
itself. Restored the original ffdelay script's naming convention
(_delayed500ms / _delayed-500ms) so the applied offset is visible in
the filename, matching what the CLI tool already did before this GUI
existed.

.gitignore

@@ -2,3 +2,4 @@ __pycache__/
 *.pyc
 .venv/
 test_media/seek_validation/
+test_media/*_fixed*

README.md

@@ -0,0 +1,4 @@
+# WeAreSyncing
+Fix audio synchronization by detecting and marking natural 'claps' in the video, or automatically determine the offset using the difference between the audio and video stream ends.
+
+This utility was vibe coded with Claude Code AI, proceed at your own discretion.

WeAreSyncing.desktop

@@ -8,3 +8,4 @@ Terminal=false
 Categories=AudioVideo;Video;
 MimeType=video/mp4;video/quicktime;video/x-matroska;video/x-msvideo;video/webm;video/mpeg;
 StartupNotify=true
+Name[en_US]=WeAreSyncing

clap_detection.py

@@ -5,6 +5,7 @@ paths use one implementation instead of drifting apart.
 import os
 import subprocess
 import tempfile
+import warnings
 from dataclasses import dataclass
 from typing import List, Optional
 
@@ -18,16 +19,22 @@ class ClapDetectionUnavailable(RuntimeError):
     pass
 
 
+# Shared with the GUI's detection-settings dialog (main.py) so both the CLI
+# and GUI default to, and can be reset to, the exact same values.
+DEFAULT_THRESHOLD = 0.7
+DEFAULT_MIN_INTERVAL = 0.3
+
+
 @dataclass(frozen=True)
 class ClapEvent:
     time_sec: float
     strength: float
 
 
-def _probe_audio_start_time(input_path: str) -> float:
+def _probe_audio_start_time(input_path: str, stream_specifier: str) -> Optional[float]:
     result = subprocess.run(
         [
-            "ffprobe", "-v", "error", "-select_streams", "a:0",
+            "ffprobe", "-v", "error", "-select_streams", stream_specifier,
             "-show_entries", "stream=start_time",
             "-of", "default=noprint_wrappers=1:nokey=1",
             input_path,
@@ -37,10 +44,20 @@ def _probe_audio_start_time(input_path: str) -> float:
     try:
         return float(result.stdout.strip())
     except ValueError:
-        return 0.0
+        # Some containers/codecs legitimately report "N/A" (genuinely no
+        # start offset to speak of), but this could also mean ffprobe just
+        # couldn't express a real nonzero offset for this container -- we
+        # can't tell those apart here, so surface it instead of silently
+        # assuming 0, which would leave onset times subtly, silently wrong.
+        warnings.warn(
+            f"could not determine audio start_time for {input_path!r} "
+            f"(ffprobe returned {result.stdout.strip()!r}); assuming 0.0, "
+            "which may be wrong for this file's container/codec."
+        )
+        return None
 
 
-def _extract_audio_wav(input_path: str) -> str:
+def _extract_audio_wav(input_path: str, audio_stream_index: Optional[int] = None) -> str:
     # The audio stream's own start_time is exactly how a container encodes
     # "this track starts playing N seconds later than the others" -- i.e.
     # it's the very thing this whole tool exists to fix. ffmpeg's default
@@ -48,14 +65,44 @@ def _extract_audio_wav(input_path: str) -> str:
     # which would silently make onset detection blind to that offset.
     # Compensate by inserting the missing lead-in as real silence, so times
     # measured from this WAV's t=0 still match the file's true timeline.
-    audio_start = _probe_audio_start_time(input_path)
+    #
+    # audio_stream_index (ffprobe/ffmpeg global stream index, e.g. mpv's
+    # own `ff-index` for the track it's currently playing) makes stream
+    # selection explicit and -- critically -- lets the caller guarantee
+    # this matches whichever audio track is actually audible in the
+    # player. Without it, ffmpeg's own default-stream heuristic picks
+    # *a* audio track, which isn't guaranteed to be the same one on
+    # multi-audio-track files (extra language dubs, a separate mic/system
+    # audio track from a screen recorder, etc.) -- something our own
+    # single-track synthetic test fixtures never exercise, but real-world
+    # files often have.
+    stream_specifier = f"{audio_stream_index}" if audio_stream_index is not None else "a:0"
+    audio_start = _probe_audio_start_time(input_path, stream_specifier) or 0.0
 
     fd, wav_path = tempfile.mkstemp(suffix=".wav")
     os.close(fd)
     cmd = ["ffmpeg", "-y", "-i", input_path, "-vn"]
+    cmd += ["-map", f"0:{audio_stream_index}" if audio_stream_index is not None else "a:0"]
+    # aresample=async=1 keeps the decoded audio conformed to its own
+    # internal PTS, filling any gap with silence instead of silently
+    # closing it up. Long recordings (stream captures, OBS-style long
+    # sessions) can have a genuine multi-second discontinuity mid-file --
+    # e.g. a brief audio-device dropout -- from a real capture glitch, not
+    # anything to do with sync between tracks. Plain -vn extraction just
+    # concatenates packets and ignores the gap, silently shortening the
+    # output and shifting every onset after that point earlier by however
+    # long the gap was. Confirmed on a real ~110min capture with exactly
+    # one such gap (+1.621s at ~15.5min in): without this filter the
+    # extracted WAV came out 1.6s short of the source's true duration;
+    # with it, the duration matches exactly. This is a different failure
+    # mode from the leading start_time offset below (which aresample alone
+    # does NOT fix -- it re-zeros to its own first frame, same as plain
+    # extraction), so both are needed together, not either/or.
+    af_chain = ["aresample=async=1"]
     if audio_start > 0.001:
         delay_ms = int(round(audio_start * 1000))
-        cmd += ["-af", f"adelay={delay_ms}|{delay_ms}"]
+        af_chain.append(f"adelay={delay_ms}|{delay_ms}")
+    cmd += ["-af", ",".join(af_chain)]
     cmd += ["-acodec", "pcm_s16le", "-ar", "22050", "-ac", "1", wav_path]
 
     result = subprocess.run(cmd, capture_output=True, text=True)
@@ -68,14 +115,19 @@ def _extract_audio_wav(input_path: str) -> str:
 def detect_claps(
     input_path: str,
     start_offset: float = 0.0,
-    threshold: float = 0.7,
-    min_interval: float = 0.3,
+    threshold: float = DEFAULT_THRESHOLD,
+    min_interval: float = DEFAULT_MIN_INTERVAL,
     max_results: Optional[int] = None,
+    audio_stream_index: Optional[int] = None,
 ) -> List[ClapEvent]:
     """Detect sharp audio onsets in input_path. `start_offset`: only onsets
     at or after this absolute timestamp are returned (for a from-cursor
     search). `max_results` caps how many are returned, for a "find the next
-    one" search instead of a full scan.
+    one" search instead of a full scan. `audio_stream_index` (an ffmpeg
+    global stream index) pins detection to a specific audio track on
+    multi-audio-track files -- pass the ff-index of whichever track mpv is
+    actually playing (see main.py) so detection can't silently analyze a
+    different track than the one being listened to.
 
     The whole file is always decoded and analyzed regardless of
     start_offset -- an earlier version tried to skip ahead with ffmpeg's
@@ -92,7 +144,7 @@ def detect_claps(
             "librosa is not installed. Install with: pip install librosa numpy soundfile"
         )
 
-    wav_path = _extract_audio_wav(input_path)
+    wav_path = _extract_audio_wav(input_path, audio_stream_index=audio_stream_index)
     try:
         y, sr = librosa.load(wav_path, sr=22050)
         onset_env = librosa.onset.onset_strength(y=y, sr=sr)

main.py

@@ -1,4 +1,6 @@
+import os
 import sys
+import tempfile
 from pathlib import Path
 from typing import Optional
 
@@ -6,53 +8,87 @@ from PyQt6.QtCore import Qt, QTimer, QThread, pyqtSignal
 from PyQt6.QtGui import QShortcut, QKeySequence, QAction
 from PyQt6.QtWidgets import (
     QApplication, QMainWindow, QWidget, QVBoxLayout, QHBoxLayout,
-    QPushButton, QLabel, QSlider, QCheckBox, QDoubleSpinBox, QFileDialog,
-    QMessageBox, QPlainTextEdit, QGroupBox,
+    QPushButton, QLabel, QDoubleSpinBox, QFileDialog,
+    QMessageBox, QPlainTextEdit, QGroupBox, QTabWidget, QDialog, QDialogButtonBox, QFormLayout,
 )
 
-from player_widget import PlayerWidget
+from player_panel import PlayerPanel, format_timestamp
 from sync_logic import compute_offset, SyncOffset
 from ffmpeg_export import export_synced, FfmpegExportError, ExportResult
-from clap_detection import detect_claps, ClapDetectionUnavailable, ClapEvent
-from marker_strip import MarkerStrip
+from clap_detection import (
+    detect_claps, ClapDetectionUnavailable, ClapEvent,
+    DEFAULT_THRESHOLD, DEFAULT_MIN_INTERVAL,
+)
+from stream_edges import probe_stream_edges, StreamEdges, StreamEdgesError
 
 
-def format_timestamp(seconds: float) -> str:
-    m, s = divmod(max(seconds, 0.0), 60)
-    return f"{int(m):02d}:{s:06.3f}"
+class ClapSettingsDialog(QDialog):
+    """Lets the threshold/min-interval knobs from clap_detection.py be
+    tuned from the GUI instead of only via the CLI's -t/-m flags. Keeps its
+    own pending values so Cancel doesn't touch the caller's settings."""
 
+    def __init__(self, threshold: float, min_interval: float, parent=None):
+        super().__init__(parent)
+        self.setWindowTitle("Clap Detection Settings")
 
-class ClickableSlider(QSlider):
-    """QSlider normally only jumps to the clicked position if you click
-    exactly on the handle; clicking elsewhere in the groove pages by a
-    fixed step. Override so any click along the track jumps straight
-    there, then let the base class take over so dragging still works
-    (the handle is now under the cursor, so its own drag tracking picks
-    up seamlessly).
-    """
+        self.threshold_spin = QDoubleSpinBox()
+        self.threshold_spin.setRange(0.1, 2.0)
+        self.threshold_spin.setSingleStep(0.05)
+        self.threshold_spin.setDecimals(2)
+        self.threshold_spin.setValue(threshold)
 
-    def mousePressEvent(self, event):
-        if event.button() == Qt.MouseButton.LeftButton and self.maximum() > self.minimum():
-            ratio = min(max(event.position().x() / max(self.width(), 1), 0.0), 1.0)
-            value = round(self.minimum() + ratio * (self.maximum() - self.minimum()))
-            self.setValue(value)
-            self.sliderMoved.emit(value)
-        super().mousePressEvent(event)
+        self.min_interval_spin = QDoubleSpinBox()
+        self.min_interval_spin.setRange(0.05, 5.0)
+        self.min_interval_spin.setSingleStep(0.05)
+        self.min_interval_spin.setDecimals(2)
+        self.min_interval_spin.setValue(min_interval)
+
+        form = QFormLayout()
+        form.addRow("Sensitivity threshold (higher = less sensitive):", self.threshold_spin)
+        form.addRow("Minimum seconds between claps:", self.min_interval_spin)
+
+        reset_button = QPushButton("Reset to Defaults")
+        reset_button.clicked.connect(self._reset_to_defaults)
+
+        buttons = QDialogButtonBox(QDialogButtonBox.StandardButton.Ok | QDialogButtonBox.StandardButton.Cancel)
+        buttons.accepted.connect(self.accept)
+        buttons.rejected.connect(self.reject)
+
+        button_row = QHBoxLayout()
+        button_row.addWidget(reset_button)
+        button_row.addStretch(1)
+        button_row.addWidget(buttons)
+
+        layout = QVBoxLayout(self)
+        layout.addLayout(form)
+        layout.addLayout(button_row)
+
+    def _reset_to_defaults(self):
+        self.threshold_spin.setValue(DEFAULT_THRESHOLD)
+        self.min_interval_spin.setValue(DEFAULT_MIN_INTERVAL)
+
+    def values(self):
+        return self.threshold_spin.value(), self.min_interval_spin.value()
 
 
 class ExportWorker(QThread):
     finished_ok = pyqtSignal(object)   # ExportResult
     finished_err = pyqtSignal(str, str)  # message, stderr
 
-    def __init__(self, input_path, delay_ms, output_path):
+    def __init__(self, input_path, delay_ms, output_path, clip_start=None, clip_duration=None):
         super().__init__()
         self.input_path = input_path
         self.delay_ms = delay_ms
         self.output_path = output_path
+        self.clip_start = clip_start
+        self.clip_duration = clip_duration
 
     def run(self):
         try:
-            result: ExportResult = export_synced(self.input_path, self.delay_ms, self.output_path)
+            result: ExportResult = export_synced(
+                self.input_path, self.delay_ms, self.output_path,
+                clip_start=self.clip_start, clip_duration=self.clip_duration,
+            )
             self.finished_ok.emit(result)
         except FfmpegExportError as e:
             self.finished_err.emit(str(e), e.stderr)
@@ -62,11 +98,15 @@ class DetectWorker(QThread):
     finished_ok = pyqtSignal(list)   # List[ClapEvent]
     finished_err = pyqtSignal(str)
 
-    def __init__(self, input_path, start_offset=0.0, max_results=None):
+    def __init__(self, input_path, start_offset=0.0, max_results=None, audio_stream_index=None,
+                 threshold=DEFAULT_THRESHOLD, min_interval=DEFAULT_MIN_INTERVAL):
         super().__init__()
         self.input_path = input_path
         self.start_offset = start_offset
         self.max_results = max_results
+        self.audio_stream_index = audio_stream_index
+        self.threshold = threshold
+        self.min_interval = min_interval
 
     def run(self):
         try:
@@ -74,30 +114,66 @@ class DetectWorker(QThread):
                 self.input_path,
                 start_offset=self.start_offset,
                 max_results=self.max_results,
+                audio_stream_index=self.audio_stream_index,
+                threshold=self.threshold,
+                min_interval=self.min_interval,
             )
             self.finished_ok.emit(events)
         except (ClapDetectionUnavailable, RuntimeError) as e:
             self.finished_err.emit(str(e))
 
 
+class EstimateWorker(QThread):
+    finished_ok = pyqtSignal(object)   # StreamEdges
+    finished_err = pyqtSignal(str)
+
+    def __init__(self, input_path, video_stream_index=None, audio_stream_index=None):
+        super().__init__()
+        self.input_path = input_path
+        self.video_stream_index = video_stream_index
+        self.audio_stream_index = audio_stream_index
+
+    def run(self):
+        try:
+            edges = probe_stream_edges(
+                self.input_path,
+                video_stream_index=self.video_stream_index,
+                audio_stream_index=self.audio_stream_index,
+            )
+            self.finished_ok.emit(edges)
+        except StreamEdgesError as e:
+            self.finished_err.emit(str(e))
+
+
 class MainWindow(QMainWindow):
     def __init__(self):
         super().__init__()
-        self.setWindowTitle("A/V Sync Marker")
+        self.setWindowTitle("WeAreSyncing")
         self.resize(960, 720)
 
         self.input_path: Optional[str] = None
         self.video_mark: Optional[float] = None
         self.audio_mark: Optional[float] = None
+        # The offset currently feeding Preview/Export -- derived either from
+        # the two markers or from the stream-ends estimate, whichever was
+        # applied last.
         self.offset: Optional[SyncOffset] = None
         self._export_worker: Optional[ExportWorker] = None
         self._detect_worker: Optional[DetectWorker] = None
-        self._slider_being_dragged = False
+        self._estimate_worker: Optional[EstimateWorker] = None
+        self._preview_worker: Optional[ExportWorker] = None
+        self._preview_temp_path: Optional[str] = None  # currently loaded in preview_panel
+        self._pending_preview_path: Optional[str] = None  # being built by _preview_worker
+        self.clap_threshold: float = DEFAULT_THRESHOLD
+        self.clap_min_interval: float = DEFAULT_MIN_INTERVAL
 
-        self.player = PlayerWidget()
-        self.player.setMinimumHeight(400)
-        self.player.fileDropped.connect(self._load)
-        self.player.rightClicked.connect(self._toggle_play)
+        # Two independent monitors (Kdenlive-style project/clip monitor
+        # tabs): each is a full player with its own timeline and transport,
+        # not just a shared video frame with one set of controls.
+        self.source_panel = PlayerPanel(show_marker_strip=True)
+        self.source_panel.fileDropped.connect(self._load)
+
+        self.preview_panel = PlayerPanel(show_marker_strip=False)
 
         self._build_ui()
         self._build_menu()
@@ -110,86 +186,67 @@ class MainWindow(QMainWindow):
 
     # -- UI construction ---------------------------------------------------
 
+    def _active_panel(self) -> PlayerPanel:
+        return self.source_panel if self.video_tabs.currentIndex() == 0 else self.preview_panel
+
     def _build_ui(self):
         central = QWidget()
         layout = QVBoxLayout(central)
 
-        layout.addWidget(self.player, stretch=1)
-
-        # Timeline directly under the video (LosslessCut-style), transport
-        # controls below that.
-        seek_row = QHBoxLayout()
-        self.position_label = QLabel("00:00.000")
-        seek_row.addWidget(self.position_label)
-
-        slider_stack = QVBoxLayout()
-        slider_stack.setSpacing(0)
-        self.marker_strip = MarkerStrip()
-        self.marker_strip.markerClicked.connect(self._on_marker_clicked)
-        slider_stack.addWidget(self.marker_strip)
-
-        self.seek_slider = ClickableSlider(Qt.Orientation.Horizontal)
-        self.seek_slider.setRange(0, 0)
-        self.seek_slider.sliderPressed.connect(lambda: setattr(self, "_slider_being_dragged", True))
-        self.seek_slider.sliderMoved.connect(self._on_slider_moved)
-        self.seek_slider.sliderReleased.connect(self._on_slider_released)
-        slider_stack.addWidget(self.seek_slider)
-        seek_row.addLayout(slider_stack, stretch=1)
-
-        self.duration_label = QLabel("00:00.000")
-        seek_row.addWidget(self.duration_label)
-        layout.addLayout(seek_row)
-
-        # Transport row: frame-step/play controls centered as a group, mute
-        # pinned to the right. Load File now lives in the File menu.
-        transport = QHBoxLayout()
-        transport.addStretch(1)
-
-        self.prev_frame_button = QPushButton("⏮ Frame (,)")
-        self.prev_frame_button.clicked.connect(lambda: self._step_frame(-1))
-        transport.addWidget(self.prev_frame_button)
-
-        self.play_button = QPushButton("Play")
-        self.play_button.clicked.connect(self._toggle_play)
-        transport.addWidget(self.play_button)
-
-        self.next_frame_button = QPushButton("Frame ⏭ (.)")
-        self.next_frame_button.clicked.connect(lambda: self._step_frame(1))
-        transport.addWidget(self.next_frame_button)
-
-        transport.addStretch(1)
-
-        self.mute_checkbox = QCheckBox("Mute audio")
-        self.mute_checkbox.toggled.connect(self._on_mute_toggled)
-        transport.addWidget(self.mute_checkbox)
-
-        layout.addLayout(transport)
+        self.video_tabs = QTabWidget()
+        self.video_tabs.addTab(self.source_panel, "Source")
+        self.video_tabs.addTab(self.preview_panel, "Preview")
+        layout.addWidget(self.video_tabs, stretch=1)
 
         # Marker row
         marker_group = QGroupBox("Sync markers")
         marker_layout = QHBoxLayout(marker_group)
 
-        self.mark_video_button = QPushButton("Mark Video Event")
-        self.mark_video_button.clicked.connect(self._mark_video)
-        marker_layout.addWidget(self.mark_video_button)
-        self.video_mark_label = QLabel("video: —")
-        marker_layout.addWidget(self.video_mark_label)
-
         self.mark_audio_button = QPushButton("Mark Audio Event")
         self.mark_audio_button.clicked.connect(self._mark_audio)
         marker_layout.addWidget(self.mark_audio_button)
         self.audio_mark_label = QLabel("audio: —")
         marker_layout.addWidget(self.audio_mark_label)
 
+        self.mark_video_button = QPushButton("Mark Video Event")
+        self.mark_video_button.clicked.connect(self._mark_video)
+        marker_layout.addWidget(self.mark_video_button)
+        self.video_mark_label = QLabel("video: —")
+        marker_layout.addWidget(self.video_mark_label)
+
+        # Re-applies the marker-derived offset after the stream-ends
+        # estimate replaced it, without having to re-mark an event.
+        self.use_marker_offset_button = QPushButton("Use Marker Offset")
+        self.use_marker_offset_button.setEnabled(False)
+        self.use_marker_offset_button.clicked.connect(self._recompute_offset)
+        marker_layout.addWidget(self.use_marker_offset_button)
+
+        marker_layout.addStretch(1)
+
+        self.detect_all_button = QPushButton("Detect All Claps")
+        self.detect_all_button.setEnabled(False)
+        self.detect_all_button.clicked.connect(self._on_detect_all_clicked)
+        marker_layout.addWidget(self.detect_all_button)
+
+        self.detect_next_button = QPushButton("Detect Next Clap")
+        self.detect_next_button.setEnabled(False)
+        self.detect_next_button.clicked.connect(self._on_detect_next_clicked)
+        marker_layout.addWidget(self.detect_next_button)
+
         layout.addWidget(marker_group)
 
-        # Offset + export row
-        export_group = QGroupBox("Offset and export")
+        # Offset + preview + export row
+        export_group = QGroupBox("Offset, preview and export")
         export_layout = QHBoxLayout(export_group)
 
         self.offset_label = QLabel("Offset: — (mark both events)")
         export_layout.addWidget(self.offset_label, stretch=1)
 
+        self.estimate_button = QPushButton("Estimate from Stream Ends")
+        self.estimate_button.setEnabled(False)
+        self.estimate_button.clicked.connect(self._on_estimate_clicked)
+        export_layout.addWidget(self.estimate_button)
+
         export_layout.addWidget(QLabel("Override (ms):"))
         self.override_spinbox = QDoubleSpinBox()
         self.override_spinbox.setRange(-600000, 600000)
@@ -197,6 +254,11 @@ class MainWindow(QMainWindow):
         self.override_spinbox.setEnabled(False)
         export_layout.addWidget(self.override_spinbox)
 
+        self.preview_button = QPushButton("Preview Sync…")
+        self.preview_button.setEnabled(False)
+        self.preview_button.clicked.connect(self._on_preview_clicked)
+        export_layout.addWidget(self.preview_button)
+
         self.export_button = QPushButton("Export…")
         self.export_button.setEnabled(False)
         self.export_button.clicked.connect(self._on_export_clicked)
@@ -238,14 +300,40 @@ class MainWindow(QMainWindow):
         self.detect_next_action.triggered.connect(self._on_detect_next_clicked)
         tools_menu.addAction(self.detect_next_action)
 
+        tools_menu.addSeparator()
+
+        self.estimate_action = QAction("Estimate Offset from Stream Ends", self)
+        self.estimate_action.setEnabled(False)
+        self.estimate_action.triggered.connect(self._on_estimate_clicked)
+        tools_menu.addAction(self.estimate_action)
+
+        tools_menu.addSeparator()
+
+        settings_action = QAction("Clap Detection Settings…", self)
+        settings_action.triggered.connect(self._on_clap_settings_clicked)
+        tools_menu.addAction(settings_action)
+
+    def _on_clap_settings_clicked(self):
+        dialog = ClapSettingsDialog(self.clap_threshold, self.clap_min_interval, parent=self)
+        if dialog.exec() == QDialog.DialogCode.Accepted:
+            self.clap_threshold, self.clap_min_interval = dialog.values()
+            self.statusBar().showMessage(
+                f"Clap detection settings updated: threshold={self.clap_threshold:.2f}, "
+                f"min_interval={self.clap_min_interval:.2f}s", 4000,
+            )
+
     def _wire_shortcuts(self):
-        QShortcut(QKeySequence(","), self, activated=lambda: self._step_frame(-1))
-        QShortcut(QKeySequence("."), self, activated=lambda: self._step_frame(1))
-        QShortcut(QKeySequence(Qt.Key.Key_Space), self, activated=self._toggle_play)
+        # These act on whichever tab (Source/Preview) is currently active --
+        # each is a fully independent monitor with its own transport, so the
+        # keyboard shortcuts follow the same "whichever one you're looking
+        # at" rule as the on-screen buttons.
+        QShortcut(QKeySequence(","), self, activated=lambda: self._active_panel().step_frame(-1))
+        QShortcut(QKeySequence("."), self, activated=lambda: self._active_panel().step_frame(1))
+        QShortcut(QKeySequence(Qt.Key.Key_Space), self, activated=lambda: self._active_panel().toggle_play())
         # QShortcut auto-repeats by default while the key is held, so this
         # keeps skipping for as long as the arrow key is down.
-        QShortcut(QKeySequence(Qt.Key.Key_Left), self, activated=lambda: self._seek_relative(-5.0))
-        QShortcut(QKeySequence(Qt.Key.Key_Right), self, activated=lambda: self._seek_relative(5.0))
+        QShortcut(QKeySequence(Qt.Key.Key_Left), self, activated=lambda: self._active_panel().seek_relative(-5.0))
+        QShortcut(QKeySequence(Qt.Key.Key_Right), self, activated=lambda: self._active_panel().seek_relative(5.0))
 
     # -- transport -----------------------------------------------------
 
@@ -255,15 +343,15 @@ class MainWindow(QMainWindow):
             self._load(path)
 
     def _load(self, path: str):
-        # player.load() blocks the GUI thread until mpv reports the file's
-        # duration, so show the status message and force one repaint before
-        # starting it -- otherwise Qt wouldn't paint the message until after
-        # the blocking call already returned.
+        # source_panel.load() blocks the GUI thread until mpv reports the
+        # file's duration, so show the status message and force one repaint
+        # before starting it -- otherwise Qt wouldn't paint the message
+        # until after the blocking call already returned.
         self.statusBar().showMessage(f"Loading {Path(path).name}…")
         QApplication.processEvents()
 
         self.input_path = path
-        self.player.load(path)
+        self.source_panel.load(path)
         self.video_mark = None
         self.audio_mark = None
         self.offset = None
@@ -271,77 +359,39 @@ class MainWindow(QMainWindow):
         self.audio_mark_label.setText("audio: —")
         self.offset_label.setText("Offset: — (mark both events)")
         self.export_button.setEnabled(False)
+        self.preview_button.setEnabled(False)
         self.override_spinbox.setEnabled(False)
+        self.use_marker_offset_button.setEnabled(False)
         self.log_view.clear()
 
-        duration = self.player.mpv.duration or 0.0
-        self.seek_slider.setRange(0, int(duration * 1000))
-        self.duration_label.setText(format_timestamp(duration))
-        self.setWindowTitle(f"A/V Sync Marker — {Path(path).name}")
-
-        self.marker_strip.set_duration(duration)
-        self.marker_strip.clear()
+        self.setWindowTitle(f"WeAreSyncing — {Path(path).name}")
         self.statusBar().showMessage(f"Loaded {Path(path).name}", 4000)
         self.detect_all_action.setEnabled(True)
+        self.detect_all_button.setEnabled(True)
         self.detect_next_action.setEnabled(True)
+        self.detect_next_button.setEnabled(True)
+        self.estimate_action.setEnabled(True)
+        self.estimate_button.setEnabled(True)
 
-    def _toggle_play(self):
-        if self.player.mpv.pause:
-            self.player.play()
-            self.play_button.setText("Pause")
-        else:
-            self.player.pause()
-            self.play_button.setText("Play")
-
-    def _step_frame(self, n: int):
-        self.player.pause()
-        self.play_button.setText("Play")
-        self.player.step_frame(n)
-        self._update_position_ui()
-
-    def _on_mute_toggled(self, checked: bool):
-        self.player.mpv.mute = checked
-
-    def _seek_relative(self, delta_seconds: float):
-        # Unlike frame-stepping (which always pauses -- you're precisely
-        # inspecting a frame), a 5s skip is a coarse nudge while scrubbing
-        # around; seeking doesn't change mpv's pause state on its own, so
-        # just not touching play/pause here is enough to preserve whichever
-        # state the user was already in.
-        if self.input_path is None:
-            return
-        duration = self.player.mpv.duration or 0.0
-        target = min(max(self.player.get_position() + delta_seconds, 0.0), duration)
-        self.player.seek(target)
-        self._update_position_ui()
-
-    def _on_slider_moved(self, value: int):
-        self.player.seek(value / 1000.0)
-        self._update_position_ui()
-
-    def _on_slider_released(self):
-        self._slider_being_dragged = False
-        self._update_position_ui()
+        # A loaded preview belongs to the *previous* source/offset -- drop it.
+        if self._preview_temp_path and os.path.exists(self._preview_temp_path):
+            os.unlink(self._preview_temp_path)
+        self._preview_temp_path = None
+        self.video_tabs.setCurrentIndex(0)
 
     def _update_position_ui(self):
-        if self.input_path is None:
-            return
-        pos = self.player.get_position()
-        self.position_label.setText(format_timestamp(pos))
-        if not self._slider_being_dragged:
-            self.seek_slider.blockSignals(True)
-            self.seek_slider.setValue(int(pos * 1000))
-            self.seek_slider.blockSignals(False)
+        self.source_panel.update_position()
+        self.preview_panel.update_position()
 
     # -- markers / offset -----------------------------------------------
 
     def _mark_video(self):
-        self.video_mark = self.player.get_position()
+        self.video_mark = self.source_panel.player.get_position()
         self.video_mark_label.setText(f"video: {format_timestamp(self.video_mark)}")
         self._recompute_offset()
 
     def _mark_audio(self):
-        self.audio_mark = self.player.get_position()
+        self.audio_mark = self.source_panel.player.get_position()
         self.audio_mark_label.setText(f"audio: {format_timestamp(self.audio_mark)}")
         self._recompute_offset()
 
@@ -349,43 +399,196 @@ class MainWindow(QMainWindow):
         if self.video_mark is None or self.audio_mark is None:
             return
         self.offset = compute_offset(self.video_mark, self.audio_mark)
-        self.offset_label.setText(self.offset.describe())
-        self.override_spinbox.setValue(self.offset.delay_ms)
+        self._apply_offset(self.offset, self.offset.describe())
+        self.use_marker_offset_button.setEnabled(True)
+
+    def _apply_offset(self, offset: SyncOffset, label: str):
+        """Shared by both offset sources (markers, stream-ends estimate):
+        whichever is applied last is what Preview/Export use."""
+        self.offset = offset
+        self.offset_label.setText(label)
+        self.override_spinbox.setValue(offset.delay_ms)
         self.override_spinbox.setEnabled(True)
         self.export_button.setEnabled(True)
+        self.preview_button.setEnabled(True)
+
+    # -- stream-ends estimate ----------------------------------------------
+
+    def _current_video_stream_index(self) -> Optional[int]:
+        for track in self.source_panel.player.mpv.track_list:
+            if track.get("type") == "video" and track.get("selected"):
+                return track.get("ff-index")
+        return None
+
+    def _on_estimate_clicked(self):
+        if self.input_path is None:
+            return
+        self.estimate_action.setEnabled(False)
+        self.estimate_button.setEnabled(False)
+        self.statusBar().showMessage("Estimating offset from stream ends…")
+
+        self._estimate_worker = EstimateWorker(
+            self.input_path,
+            video_stream_index=self._current_video_stream_index(),
+            audio_stream_index=self._current_audio_stream_index(),
+        )
+        self._estimate_worker.finished_ok.connect(self._on_estimate_done)
+        self._estimate_worker.finished_err.connect(self._on_estimate_failed)
+        self._estimate_worker.start()
+
+    def _on_estimate_done(self, edges: StreamEdges):
+        self.estimate_action.setEnabled(True)
+        self.estimate_button.setEnabled(True)
+
+        offset = SyncOffset(delay_ms=round(edges.end_delay_ms))
+        self._apply_offset(offset, f"{offset.describe()} — estimated from stream ends")
+
+        self.log_view.appendPlainText(
+            f"Stream-ends estimate: video {edges.video_start:.3f}–{edges.video_end:.3f}s, "
+            f"audio {edges.audio_start:.3f}–{edges.audio_end:.3f}s → "
+            f"end-based {edges.end_delay_ms:+.0f} ms, start-based {edges.start_delay_ms:+.0f} ms"
+        )
+        if edges.estimates_disagree:
+            message = (
+                f"Estimated {offset.delay_ms:+.0f} ms from stream ends, but stream starts suggest "
+                f"{edges.start_delay_ms:+.0f} ms — verify with Preview."
+            )
+            self.log_view.appendPlainText("WARNING: " + message)
+        else:
+            message = f"Estimated {offset.delay_ms:+.0f} ms from stream ends (assumes both tracks stopped recording together) — verify with Preview."
+        self.statusBar().showMessage(message, 10000)
+
+    def _on_estimate_failed(self, message: str):
+        self.estimate_action.setEnabled(True)
+        self.estimate_button.setEnabled(True)
+        self.statusBar().showMessage("Offset estimate failed (see log).", 8000)
+        self.log_view.appendPlainText(f"Stream-ends estimate FAILED: {message}")
+
+    # -- preview -----------------------------------------------------
+
+    PREVIEW_MARGIN_SEC = 3.0
+    PREVIEW_DURATION_SEC = 8.0
+
+    def _on_preview_clicked(self):
+        if self.input_path is None or self.offset is None:
+            return
+        delay_ms = self.override_spinbox.value()
+
+        # Build just a short clip around the marked events rather than the
+        # whole file -- lossless stream-copy of a few seconds is fast
+        # regardless of how long the source file is, unlike previewing the
+        # full result. Centering on the midpoint of the two original marks
+        # keeps both the video and audio events (and wherever the offset
+        # moves them to) inside the window for typical human-scale marking
+        # errors.
+        #
+        # An estimated offset may have no markers behind it; center on the
+        # Source playhead instead, so the user can park it on a sync event
+        # (or any stretch of speech) before previewing.
+        if self.video_mark is not None and self.audio_mark is not None:
+            center = (self.video_mark + self.audio_mark) / 2.0
+        else:
+            center = self.source_panel.player.get_position()
+        clip_start = max(center - self.PREVIEW_MARGIN_SEC, 0.0)
+
+        input_p = Path(self.input_path)
+        fd, temp_path = tempfile.mkstemp(suffix=input_p.suffix, prefix="avsync_preview_")
+        os.close(fd)
+        self._pending_preview_path = temp_path
+
+        self.preview_button.setEnabled(False)
+        self.preview_button.setText("Building preview…")
+        self.statusBar().showMessage("Building preview clip…")
+
+        self._preview_worker = ExportWorker(
+            self.input_path, delay_ms, temp_path,
+            clip_start=f"{clip_start:.3f}", clip_duration=self.PREVIEW_DURATION_SEC,
+        )
+        self._preview_worker.finished_ok.connect(self._on_preview_ready)
+        self._preview_worker.finished_err.connect(self._on_preview_failed)
+        self._preview_worker.start()
+
+    def _on_preview_ready(self, result: ExportResult):
+        self.preview_button.setEnabled(True)
+        self.preview_button.setText("Preview Sync…")
+        self.statusBar().showMessage("Preview ready.", 4000)
+        if result.stderr.strip():
+            self.log_view.appendPlainText("ffmpeg preview log:\n" + result.stderr)
+
+        # The preview panel is long-lived (one mpv instance for the whole
+        # session), so swap files instead of spinning up a new player each
+        # time: load the new clip first, then delete the old temp file --
+        # only safe to unlink it once mpv is no longer holding it open.
+        old_path = self._preview_temp_path
+        new_path = self._pending_preview_path
+        self.preview_panel.load(new_path)
+        self.preview_panel.force_play()
+        self._preview_temp_path = new_path
+        self._pending_preview_path = None
+        self.video_tabs.setCurrentIndex(1)
+
+        if old_path and old_path != new_path and os.path.exists(old_path):
+            os.unlink(old_path)
+
+    def _on_preview_failed(self, message: str, stderr: str):
+        self.preview_button.setEnabled(True)
+        self.preview_button.setText("Preview Sync…")
+        self.statusBar().showMessage("Preview failed (see log).", 8000)
+        self.log_view.appendPlainText(f"Preview FAILED: {message}\n{stderr}")
+        if self._pending_preview_path and os.path.exists(self._pending_preview_path):
+            os.unlink(self._pending_preview_path)
+        self._pending_preview_path = None
 
     # -- clap detection -----------------------------------------------------
 
-    def _on_marker_clicked(self, time_sec: float):
-        self.player.seek(time_sec)
-        self._update_position_ui()
-
     def _on_detect_all_clicked(self):
         self._start_detection(start_offset=0.0, max_results=None, label="Scanning whole file for claps…")
 
     def _on_detect_next_clicked(self):
-        cursor = self.player.get_position()
+        cursor = self.source_panel.player.get_position()
         self._start_detection(start_offset=cursor, max_results=1, label=f"Scanning from {format_timestamp(cursor)} for next clap…")
 
+    def _current_audio_stream_index(self) -> Optional[int]:
+        """The ffmpeg/ffprobe global stream index of whichever audio track
+        mpv is currently playing in the Source panel. Passed through to
+        clap detection so it analyzes the exact track being listened to --
+        without this, ffmpeg's own default-stream selection could silently
+        pick a *different* audio track on multi-audio-track files (extra
+        language dubs, a separate mic/system-audio track from a screen
+        recorder), producing markers for the wrong track's timing entirely.
+        """
+        for track in self.source_panel.player.mpv.track_list:
+            if track.get("type") == "audio" and track.get("selected"):
+                return track.get("ff-index")
+        return None
+
     def _start_detection(self, start_offset: float, max_results, label: str):
         if self.input_path is None:
             return
         self.detect_all_action.setEnabled(False)
+        self.detect_all_button.setEnabled(False)
         self.detect_next_action.setEnabled(False)
+        self.detect_next_button.setEnabled(False)
         self.statusBar().showMessage(label)
 
-        self._detect_worker = DetectWorker(self.input_path, start_offset=start_offset, max_results=max_results)
+        self._detect_worker = DetectWorker(
+            self.input_path, start_offset=start_offset, max_results=max_results,
+            audio_stream_index=self._current_audio_stream_index(),
+            threshold=self.clap_threshold, min_interval=self.clap_min_interval,
+        )
         self._detect_worker.finished_ok.connect(self._on_detect_done)
         self._detect_worker.finished_err.connect(self._on_detect_failed)
         self._detect_worker.start()
 
     def _on_detect_done(self, events: list):
         self.detect_all_action.setEnabled(True)
+        self.detect_all_button.setEnabled(True)
         self.detect_next_action.setEnabled(True)
+        self.detect_next_button.setEnabled(True)
 
-        existing = set(self.marker_strip.markers())
+        existing = set(self.source_panel.marker_strip.markers())
         existing.update(e.time_sec for e in events)
-        self.marker_strip.set_markers(list(existing))
+        self.source_panel.marker_strip.set_markers(list(existing))
 
         if not events:
             self.statusBar().showMessage("No claps found.", 5000)
@@ -398,7 +601,9 @@ class MainWindow(QMainWindow):
 
     def _on_detect_failed(self, message: str):
         self.detect_all_action.setEnabled(True)
+        self.detect_all_button.setEnabled(True)
         self.detect_next_action.setEnabled(True)
+        self.detect_next_button.setEnabled(True)
         self.statusBar().showMessage("Detection failed (see log).", 8000)
         self.log_view.appendPlainText(f"Clap detection FAILED: {message}")
 
@@ -410,7 +615,11 @@ class MainWindow(QMainWindow):
         delay_ms = self.override_spinbox.value()
 
         input_p = Path(self.input_path)
-        suggested = str(input_p.with_name(f"{input_p.stem}_synced{input_p.suffix}"))
+        # Matches the original ffdelay script's naming convention
+        # (e.g. "_delayed500ms" / "_delayed-500ms") rather than inventing
+        # a new one.
+        delayed_str = f"{int(round(delay_ms))}"
+        suggested = str(input_p.with_name(f"{input_p.stem}_delayed{delayed_str}ms{input_p.suffix}"))
         output_path, _ = QFileDialog.getSaveFileName(self, "Export synced file", suggested)
         if not output_path:
             return
@@ -442,7 +651,15 @@ class MainWindow(QMainWindow):
         QMessageBox.critical(self, "Export failed", f"{message}\n\n{stderr[-2000:]}")
 
     def closeEvent(self, event):
-        self.player.shutdown()
+        # Stop the timer before tearing down mpv -- otherwise a tick that
+        # fires between shutdown() and the process actually exiting calls
+        # get_position() on an already-terminated core and crashes
+        # (mpv.ShutdownError, observed as a hard crash in testing).
+        self._position_timer.stop()
+        self.source_panel.shutdown()
+        self.preview_panel.shutdown()
+        if self._preview_temp_path and os.path.exists(self._preview_temp_path):
+            os.unlink(self._preview_temp_path)
         super().closeEvent(event)
 
 

player_panel.py

@@ -0,0 +1,176 @@
+from typing import Optional
+
+from PyQt6.QtCore import Qt, pyqtSignal
+from PyQt6.QtWidgets import QWidget, QVBoxLayout, QHBoxLayout, QLabel, QPushButton, QCheckBox, QSlider
+
+from player_widget import PlayerWidget
+from marker_strip import MarkerStrip
+
+
+def format_timestamp(seconds: float) -> str:
+    m, s = divmod(max(seconds, 0.0), 60)
+    return f"{int(m):02d}:{s:06.3f}"
+
+
+class ClickableSlider(QSlider):
+    """QSlider normally only jumps to the clicked position if you click
+    exactly on the handle; clicking elsewhere in the groove pages by a
+    fixed step. Override so any click along the track jumps straight
+    there, then let the base class take over so dragging still works
+    (the handle is now under the cursor, so its own drag tracking picks
+    up seamlessly).
+    """
+
+    def mousePressEvent(self, event):
+        if event.button() == Qt.MouseButton.LeftButton and self.maximum() > self.minimum():
+            ratio = min(max(event.position().x() / max(self.width(), 1), 0.0), 1.0)
+            value = round(self.minimum() + ratio * (self.maximum() - self.minimum()))
+            self.setValue(value)
+            self.sliderMoved.emit(value)
+        super().mousePressEvent(event)
+
+
+class PlayerPanel(QWidget):
+    """A self-contained video monitor: player + its own timeline + its own
+    transport row (Kdenlive-style project/clip monitor). Two of these are
+    used side by side as tabs (Source, Preview) in main.py, each with an
+    independent PlayerWidget/mpv instance and independent playback state --
+    switching tabs switches which monitor's timeline and transport you're
+    looking at, not just the video frame.
+    """
+
+    fileDropped = pyqtSignal(str)
+
+    def __init__(self, show_marker_strip: bool = False, parent=None):
+        super().__init__(parent)
+
+        self.player = PlayerWidget()
+        self.player.setMinimumHeight(320)
+        self.player.fileDropped.connect(self.fileDropped)
+        self.player.rightClicked.connect(self.toggle_play)
+
+        self._loaded = False
+        self._slider_being_dragged = False
+
+        layout = QVBoxLayout(self)
+        layout.setContentsMargins(0, 0, 0, 8)
+        layout.addWidget(self.player, stretch=1)
+
+        seek_row = QHBoxLayout()
+        self.position_label = QLabel("00:00.000")
+        seek_row.addWidget(self.position_label)
+
+        slider_stack = QVBoxLayout()
+        slider_stack.setSpacing(0)
+
+        self.marker_strip: Optional[MarkerStrip] = None
+        if show_marker_strip:
+            self.marker_strip = MarkerStrip()
+            self.marker_strip.markerClicked.connect(self._on_marker_clicked)
+            slider_stack.addWidget(self.marker_strip)
+
+        self.seek_slider = ClickableSlider(Qt.Orientation.Horizontal)
+        self.seek_slider.setRange(0, 0)
+        self.seek_slider.sliderPressed.connect(lambda: setattr(self, "_slider_being_dragged", True))
+        self.seek_slider.sliderMoved.connect(self._on_slider_moved)
+        self.seek_slider.sliderReleased.connect(self._on_slider_released)
+        slider_stack.addWidget(self.seek_slider)
+        seek_row.addLayout(slider_stack, stretch=1)
+
+        self.duration_label = QLabel("00:00.000")
+        seek_row.addWidget(self.duration_label)
+        layout.addLayout(seek_row)
+
+        transport = QHBoxLayout()
+        transport.addStretch(1)
+
+        self.prev_frame_button = QPushButton("⏮ Frame")
+        self.prev_frame_button.clicked.connect(lambda: self.step_frame(-1))
+        transport.addWidget(self.prev_frame_button)
+
+        self.play_button = QPushButton("Play")
+        self.play_button.clicked.connect(self.toggle_play)
+        transport.addWidget(self.play_button)
+
+        self.next_frame_button = QPushButton("Frame ⏭")
+        self.next_frame_button.clicked.connect(lambda: self.step_frame(1))
+        transport.addWidget(self.next_frame_button)
+
+        transport.addStretch(1)
+
+        self.mute_checkbox = QCheckBox("Mute audio")
+        self.mute_checkbox.toggled.connect(lambda checked: setattr(self.player.mpv, "mute", checked))
+        transport.addWidget(self.mute_checkbox)
+
+        layout.addLayout(transport)
+
+    # -- playback -----------------------------------------------------
+
+    def load(self, path: str):
+        self.player.load(path)
+        self._loaded = True
+        duration = self.player.mpv.duration or 0.0
+        self.seek_slider.setRange(0, int(duration * 1000))
+        self.duration_label.setText(format_timestamp(duration))
+        if self.marker_strip is not None:
+            self.marker_strip.set_duration(duration)
+            self.marker_strip.clear()
+        self.play_button.setText("Play")
+        self.update_position()
+
+    def toggle_play(self):
+        if self.player.mpv.pause:
+            self.player.play()
+            self.play_button.setText("Pause")
+        else:
+            self.player.pause()
+            self.play_button.setText("Play")
+
+    def force_play(self):
+        self.player.play()
+        self.play_button.setText("Pause")
+
+    def step_frame(self, n: int):
+        if not self._loaded:
+            return
+        self.player.pause()
+        self.play_button.setText("Play")
+        self.player.step_frame(n)
+        self.update_position()
+
+    def seek_relative(self, delta_seconds: float):
+        # Unlike frame-stepping (which always pauses -- you're precisely
+        # inspecting a frame), a coarse skip is meant to preserve whichever
+        # play/pause state the user was already in; seeking doesn't change
+        # mpv's pause state on its own, so not touching it here is enough.
+        if not self._loaded:
+            return
+        duration = self.player.mpv.duration or 0.0
+        target = min(max(self.player.get_position() + delta_seconds, 0.0), duration)
+        self.player.seek(target)
+        self.update_position()
+
+    def _on_slider_moved(self, value: int):
+        self.player.seek(value / 1000.0)
+        self.update_position()
+
+    def _on_slider_released(self):
+        self._slider_being_dragged = False
+        self.update_position()
+
+    def _on_marker_clicked(self, time_sec: float):
+        self.player.seek(time_sec)
+        self.update_position()
+
+    def update_position(self):
+        if not self._loaded:
+            return
+        pos = self.player.get_position()
+        self.position_label.setText(format_timestamp(pos))
+        if not self._slider_being_dragged:
+            self.seek_slider.blockSignals(True)
+            self.seek_slider.setValue(int(pos * 1000))
+            self.seek_slider.blockSignals(False)
+
+    def shutdown(self):
+        self.player.shutdown()

run.sh

@@ -9,5 +9,5 @@
 # /etc/ld.so.conf.d/ would fix this system-wide instead, if preferred.
 set -e
 cd "$(dirname "$0")"
-export LD_LIBRARY_PATH="/usr/local/lib:${LD_LIBRARY_PATH}"
+# export LD_LIBRARY_PATH="/usr/local/lib:${LD_LIBRARY_PATH}"
 exec .venv/bin/python3 "$@"

scripts/detect_claps.py

@@ -11,7 +11,7 @@ from datetime import timedelta
 from pathlib import Path
 
 sys.path.insert(0, str(Path(__file__).resolve().parent.parent))
-from clap_detection import detect_claps, ClapDetectionUnavailable
+from clap_detection import detect_claps, ClapDetectionUnavailable, DEFAULT_THRESHOLD, DEFAULT_MIN_INTERVAL
 
 
 def format_timestamp(seconds):
@@ -43,10 +43,10 @@ Threshold guide:
     )
 
     parser.add_argument('input_file', help='Audio or video file')
-    parser.add_argument('-t', '--threshold', type=float, default=0.7,
-                        help='Detection threshold (0-1, default: 0.7, higher = less sensitive)')
-    parser.add_argument('-m', '--min-interval', type=float, default=0.3,
-                        help='Minimum seconds between detections (default: 0.3)')
+    parser.add_argument('-t', '--threshold', type=float, default=DEFAULT_THRESHOLD,
+                        help=f'Detection threshold (0-1, default: {DEFAULT_THRESHOLD}, higher = less sensitive)')
+    parser.add_argument('-m', '--min-interval', type=float, default=DEFAULT_MIN_INTERVAL,
+                        help=f'Minimum seconds between detections (default: {DEFAULT_MIN_INTERVAL})')
     parser.add_argument('-s', '--show-strength', action='store_true',
                         help='Show onset strength values')
 

scripts/smoke_test_arrow_seek.py

@@ -51,39 +51,40 @@ def main():
     assert not win.toggle_log_action.isChecked()
     print("PASS: log panel hidden by default")
 
-    win.player.seek(5.0)
+    win.source_panel.player.seek(5.0)
     qapp.processEvents()
     QTest.keyClick(win, Qt.Key.Key_Right)
     qapp.processEvents()
     time.sleep(0.2)
     qapp.processEvents()
-    pos = win.player.get_position()
+    pos = win.source_panel.player.get_position()
     print(f"after Right arrow from 5.0s: pos={pos:.3f}s")
     assert abs(pos - 10.0) < 0.5, f"expected ~10.0s, got {pos:.3f}s"  # 10s clip, clamps at end
 
-    win.player.seek(5.0)
+    win.source_panel.player.seek(5.0)
     qapp.processEvents()
     QTest.keyClick(win, Qt.Key.Key_Left)
     qapp.processEvents()
     time.sleep(0.2)
     qapp.processEvents()
-    pos = win.player.get_position()
+    pos = win.source_panel.player.get_position()
     print(f"after Left arrow from 5.0s: pos={pos:.3f}s")
     assert abs(pos - 0.0) < 0.5, f"expected ~0.0s, got {pos:.3f}s"
 
-    win.player.seek(2.0)
+    win.source_panel.player.seek(2.0)
     qapp.processEvents()
     QTest.keyClick(win, Qt.Key.Key_Left)
     qapp.processEvents()
     time.sleep(0.2)
     qapp.processEvents()
-    pos = win.player.get_position()
+    pos = win.source_panel.player.get_position()
     print(f"after Left arrow from 2.0s (should clamp to 0, not go negative): pos={pos:.3f}s")
     assert 0.0 <= pos < 0.5, f"expected clamped to ~0.0s, got {pos:.3f}s"
 
     print("PASS: arrow keys skip 5s and clamp to file bounds")
 
-    win.player.shutdown()
+    win.source_panel.shutdown()
+    win.preview_panel.shutdown()
     print("\nARROW SEEK SMOKE TEST PASSED")
 
 

scripts/smoke_test_clap_detection.py

@@ -47,7 +47,7 @@ def main():
     ok = wait_for(lambda: win.detect_all_action.isEnabled(), qapp)
     assert ok, "detection did not finish in time"
 
-    markers = win.marker_strip.markers()
+    markers = win.source_panel.marker_strip.markers()
     print(f"detected markers: {markers}")
     assert len(markers) >= 1, "expected at least one clap marker"
     # ground truth beep is at t=5.350s in this fixture; onset detection has
@@ -59,12 +59,12 @@ def main():
     # Simulate a real mouse click on the marker to prove click-to-seek works
     # end to end, not just that the signal is wired.
     target_time = min(markers, key=lambda m: abs(m - 5.350))
-    strip = win.marker_strip
-    strip.resize(win.seek_slider.width(), strip.height())
+    strip = win.source_panel.marker_strip
+    strip.resize(win.source_panel.seek_slider.width(), strip.height())
     x = strip._x_for_time(target_time)
     click_pos = QPointF(x, 5)
 
-    before_pos = win.player.get_position()
+    before_pos = win.source_panel.player.get_position()
     event = QMouseEvent(
         QMouseEvent.Type.MouseButtonPress, click_pos, click_pos,
         Qt.MouseButton.LeftButton, Qt.MouseButton.LeftButton, Qt.KeyboardModifier.NoModifier,
@@ -74,11 +74,12 @@ def main():
     time.sleep(0.2)
     qapp.processEvents()
 
-    after_pos = win.player.get_position()
+    after_pos = win.source_panel.player.get_position()
     print(f"clicked marker at x={x:.1f} (t={target_time:.3f}s); player position before={before_pos:.3f}s after={after_pos:.3f}s")
     assert abs(after_pos - target_time) < 0.1, "clicking the marker should seek the player to that timestamp"
 
-    win.player.shutdown()
+    win.source_panel.shutdown()
+    win.preview_panel.shutdown()
     print("\nCLAP DETECTION SMOKE TEST PASSED")
 
 

scripts/smoke_test_clap_settings.py

@@ -0,0 +1,95 @@
+"""Verifies the Clap Detection Settings dialog: defaults match
+clap_detection.py's DEFAULT_THRESHOLD/DEFAULT_MIN_INTERVAL, changing values
+and accepting actually updates MainWindow state and gets used by the next
+detection run, Cancel leaves state untouched, and Reset to Defaults in the
+dialog restores the shared constants (not just whatever MainWindow happened
+to start with).
+"""
+import sys
+import time
+from pathlib import Path
+
+from PyQt6.QtWidgets import QApplication, QDialog
+
+sys.path.insert(0, str(Path(__file__).resolve().parent.parent))
+import main as app_main
+from clap_detection import DEFAULT_THRESHOLD, DEFAULT_MIN_INTERVAL
+
+ROOT = Path(__file__).resolve().parent.parent
+CLIP = str(ROOT / "test_media" / "sync_test_INSYNC.mp4")
+
+
+def wait_for(predicate, qapp, timeout=20):
+    deadline = time.time() + timeout
+    while not predicate() and time.time() < deadline:
+        qapp.processEvents()
+        time.sleep(0.05)
+    return predicate()
+
+
+def main():
+    qapp = QApplication(sys.argv)
+    win = app_main.MainWindow()
+    win.show()
+    qapp.processEvents()
+
+    assert win.clap_threshold == DEFAULT_THRESHOLD
+    assert win.clap_min_interval == DEFAULT_MIN_INTERVAL
+    print(f"PASS: MainWindow starts with defaults ({DEFAULT_THRESHOLD}, {DEFAULT_MIN_INTERVAL})")
+
+    # -- Dialog opens pre-filled with current settings
+    dialog = app_main.ClapSettingsDialog(win.clap_threshold, win.clap_min_interval)
+    assert dialog.threshold_spin.value() == DEFAULT_THRESHOLD
+    assert dialog.min_interval_spin.value() == DEFAULT_MIN_INTERVAL
+
+    # -- Changing values and cancelling shouldn't touch MainWindow state
+    dialog.threshold_spin.setValue(0.3)
+    dialog.min_interval_spin.setValue(1.5)
+    dialog.reject()
+    assert win.clap_threshold == DEFAULT_THRESHOLD, "cancel should not have changed MainWindow state"
+    print("PASS: Cancel leaves settings untouched")
+
+    # -- Changing values and accepting (simulated via the real handler) does
+    # update MainWindow state and gets used by the actual worker.
+    dialog2 = app_main.ClapSettingsDialog(win.clap_threshold, win.clap_min_interval)
+    dialog2.threshold_spin.setValue(0.3)  # very sensitive
+    dialog2.min_interval_spin.setValue(0.1)
+    win.clap_threshold, win.clap_min_interval = dialog2.values()
+    assert win.clap_threshold == 0.3
+    assert win.clap_min_interval == 0.1
+    print("PASS: Accepting new values updates MainWindow state")
+
+    # -- Reset to Defaults button restores the shared constants regardless
+    # of what was passed in (even if MainWindow itself had non-default
+    # settings already).
+    dialog3 = app_main.ClapSettingsDialog(0.3, 0.1)
+    assert dialog3.threshold_spin.value() == 0.3
+    dialog3._reset_to_defaults()
+    assert dialog3.threshold_spin.value() == DEFAULT_THRESHOLD
+    assert dialog3.min_interval_spin.value() == DEFAULT_MIN_INTERVAL
+    print("PASS: Reset to Defaults restores DEFAULT_THRESHOLD/DEFAULT_MIN_INTERVAL")
+
+    # -- End-to-end: a changed (very sensitive) threshold actually reaches
+    # the real DetectWorker/detect_claps() call for a real detection run.
+    win._load(CLIP)
+    qapp.processEvents()
+    time.sleep(0.2)
+    qapp.processEvents()
+
+    win.clap_threshold = 0.3
+    win.clap_min_interval = 0.1
+    win.detect_all_button.click()
+    qapp.processEvents()
+    assert win._detect_worker.threshold == 0.3
+    assert win._detect_worker.min_interval == 0.1
+    ok = wait_for(lambda: win.detect_all_button.isEnabled(), qapp)
+    assert ok, "detection did not finish in time"
+    print("PASS: changed settings actually reach the DetectWorker for a real run")
+
+    win.close()
+    qapp.processEvents()
+    print("\nCLAP SETTINGS SMOKE TEST PASSED")
+
+
+if __name__ == "__main__":
+    main()

scripts/smoke_test_interactions.py

@@ -30,13 +30,19 @@ def main():
     win.show()
     qapp.processEvents()
 
-    # -- Tools menu has the clap detection actions, no buttons in the layout
+    # -- Tools menu has the clap detection actions, AND both Detect All Claps
+    # and Detect Next Clap are also buttons in the Sync markers panel (each
+    # kept in sync with its menu action's enabled state, since they're two
+    # separate widgets).
     menu_titles = [a.text() for a in win.menuBar().actions()]
     assert "&Tools" in menu_titles, f"expected a Tools menu, got {menu_titles}"
-    assert not hasattr(win, "detect_all_button"), "old button attribute should be gone"
     assert win.detect_all_action.text() == "Detect All Claps"
     assert win.detect_next_action.text() == "Detect Next Clap"
-    print("PASS: Tools menu has plain-labeled clap detection actions")
+    assert win.detect_all_button.text() == "Detect All Claps"
+    assert win.detect_next_button.text() == "Detect Next Clap"
+    assert win.detect_all_action.isEnabled() == win.detect_all_button.isEnabled()
+    assert win.detect_next_action.isEnabled() == win.detect_next_button.isEnabled()
+    print("PASS: Tools menu has plain-labeled clap detection actions, both buttons back in sync markers panel")
 
     # -- Load via drag-and-drop simulation on the player widget
     mime = QMimeData()
@@ -46,7 +52,7 @@ def main():
         Qt.MouseButton.LeftButton, Qt.KeyboardModifier.NoModifier,
         QDropEvent.Type.Drop,
     )
-    win.player.dropEvent(drop_event)
+    win.source_panel.player.dropEvent(drop_event)
     qapp.processEvents()
     time.sleep(0.3)
     qapp.processEvents()
@@ -54,42 +60,43 @@ def main():
     print(f"PASS: drag-and-drop loaded {Path(win.input_path).name}")
 
     # -- Clicking directly on the slider track (not the handle) jumps there
-    duration_ms = win.seek_slider.maximum()
+    duration_ms = win.source_panel.seek_slider.maximum()
     assert duration_ms > 0
-    target_x = win.seek_slider.width() * 0.5
-    click_pos = QPointF(target_x, win.seek_slider.height() / 2)
+    target_x = win.source_panel.seek_slider.width() * 0.5
+    click_pos = QPointF(target_x, win.source_panel.seek_slider.height() / 2)
     press_event = QMouseEvent(
         QMouseEvent.Type.MouseButtonPress, click_pos, click_pos,
         Qt.MouseButton.LeftButton, Qt.MouseButton.LeftButton, Qt.KeyboardModifier.NoModifier,
     )
-    before = win.player.get_position()
-    win.seek_slider.mousePressEvent(press_event)
+    before = win.source_panel.player.get_position()
+    win.source_panel.seek_slider.mousePressEvent(press_event)
     qapp.processEvents()
     time.sleep(0.2)
     qapp.processEvents()
-    after = win.player.get_position()
+    after = win.source_panel.player.get_position()
     expected = 0.5 * (duration_ms / 1000.0)
     print(f"clicked slider track at 50%: before={before:.3f}s after={after:.3f}s expected~{expected:.3f}s")
     assert abs(after - expected) < 0.5, "clicking the track should jump the player near that position"
     print("PASS: click-anywhere-on-track seeking works")
 
     # -- Dragging (sliderMoved) seeks live, without waiting for release
-    win.seek_slider.sliderPressed.emit()
+    win.source_panel.seek_slider.sliderPressed.emit()
     qapp.processEvents()
     drag_value = int(0.2 * duration_ms)
-    win.seek_slider.setValue(drag_value)
-    win.seek_slider.sliderMoved.emit(drag_value)
+    win.source_panel.seek_slider.setValue(drag_value)
+    win.source_panel.seek_slider.sliderMoved.emit(drag_value)
     qapp.processEvents()
     time.sleep(0.2)
     qapp.processEvents()
-    live_pos = win.player.get_position()
+    live_pos = win.source_panel.player.get_position()
     expected_drag = drag_value / 1000.0
     print(f"dragged to 20% (still 'pressed', not released): live player position={live_pos:.3f}s expected~{expected_drag:.3f}s")
     assert abs(live_pos - expected_drag) < 0.5, "dragging should seek immediately, before release"
-    win.seek_slider.sliderReleased.emit()
+    win.source_panel.seek_slider.sliderReleased.emit()
     print("PASS: live-seek during drag works")
 
-    win.player.shutdown()
+    win.source_panel.shutdown()
+    win.preview_panel.shutdown()
     print("\nALL INTERACTION SMOKE TESTS PASSED")
 
 

scripts/smoke_test_multitrack_clap_detection.py

@@ -0,0 +1,88 @@
+"""Verifies clap detection analyzes whichever audio track mpv is actually
+playing, not whichever one ffmpeg's own default-stream heuristic would
+pick. Uses a synthetic fixture with two audio tracks with claps at very
+different times (3s vs 7s) -- if detection ignored mpv's active track
+selection, switching mpv to the second track would still incorrectly
+report the first track's 3s clap instead of the second track's 7s one.
+
+This directly tests the bug hypothesis from user report: 'sometimes too
+early, sometimes too late, by different amounts' -- exactly what you'd
+see if detection sometimes analyzed the wrong audio track on
+multi-audio-track files, since a wrong-track clap position is essentially
+unrelated to the true one (not a small timestamp-math error).
+"""
+import sys
+import time
+from pathlib import Path
+
+from PyQt6.QtWidgets import QApplication
+
+sys.path.insert(0, str(Path(__file__).resolve().parent.parent))
+import main as app_main
+
+ROOT = Path(__file__).resolve().parent.parent
+CLIP = str(ROOT / "test_media" / "sync_test_MULTITRACK.mp4")
+
+
+def wait_for(predicate, qapp, timeout=20):
+    deadline = time.time() + timeout
+    while not predicate() and time.time() < deadline:
+        qapp.processEvents()
+        time.sleep(0.05)
+    return predicate()
+
+
+def main():
+    qapp = QApplication(sys.argv)
+    win = app_main.MainWindow()
+    win.show()
+    qapp.processEvents()
+
+    win._load(CLIP)
+    qapp.processEvents()
+    time.sleep(0.3)
+    qapp.processEvents()
+
+    mpv = win.source_panel.player.mpv
+    print(f"default aid: {mpv.aid}")
+    default_index = win._current_audio_stream_index()
+    print(f"resolved ff-index for default track: {default_index}")
+    assert default_index == 1, f"expected default track's ff-index to be 1, got {default_index}"
+
+    win.detect_all_button.click()
+    qapp.processEvents()
+    ok = wait_for(lambda: win.detect_all_button.isEnabled(), qapp)
+    assert ok, "detection did not finish in time"
+    markers = win.source_panel.marker_strip.markers()
+    print(f"markers with default track selected: {markers}")
+    assert any(abs(m - 3.0) < 0.5 for m in markers), f"expected a marker near 3.0s (default track's clap): {markers}"
+    assert not any(abs(m - 7.0) < 0.5 for m in markers), f"should not see the OTHER track's 7.0s clap yet: {markers}"
+
+    # Now switch mpv to the second audio track (aid=2), simulating a file
+    # where the user is listening to a non-default track, and rerun
+    # detection -- it must follow the switch, not keep reporting the first
+    # track's clap.
+    mpv.aid = 2
+    qapp.processEvents()
+    time.sleep(0.2)
+    switched_index = win._current_audio_stream_index()
+    print(f"resolved ff-index after switching to aid=2: {switched_index}")
+    assert switched_index == 2, f"expected ff-index 2 after switching tracks, got {switched_index}"
+
+    win.source_panel.marker_strip.clear()
+    win.detect_all_button.click()
+    qapp.processEvents()
+    ok = wait_for(lambda: win.detect_all_button.isEnabled(), qapp)
+    assert ok, "detection did not finish in time"
+    markers = win.source_panel.marker_strip.markers()
+    print(f"markers with second track selected: {markers}")
+    assert any(abs(m - 7.0) < 0.5 for m in markers), f"expected a marker near 7.0s (second track's clap) after switching: {markers}"
+    assert not any(abs(m - 3.0) < 0.5 for m in markers), f"should not see the FIRST track's 3.0s clap after switching: {markers}"
+
+    win.close()
+    qapp.processEvents()
+    print("\nPASS: clap detection follows mpv's actual selected audio track, not ffmpeg's default")
+
+
+if __name__ == "__main__":
+    main()

scripts/smoke_test_preserve_play_state.py

@@ -1,14 +1,14 @@
 """Verifies arrow-key skips preserve whichever play/pause state the user
 was already in, instead of always forcing a pause.
 
-Calls the real MainWindow._seek_relative() method directly rather than
+Calls the real PlayerPanel.seek_relative() method directly rather than
 simulating the QShortcut keypress: QTest.keyClick against this shortcut was
 observed to be unreliable in this headless setup even with Qt's
 active-window state forced (see smoke_test_arrow_seek.py's note) --
 sometimes the simulated key silently doesn't reach the shortcut at all, and
 a debug script isolated that PlayerWidget.seek() itself works correctly
 while playing, so the flakiness is in the synthetic-keypress dispatch, not
-the app. The QShortcut -> _seek_relative wiring itself (a single line) was
+the app. The QShortcut -> seek_relative wiring itself (a single line) was
 already independently confirmed with real xdotool keypresses against a
 live window in an earlier session.
 """
@@ -37,37 +37,38 @@ def main():
     qapp.processEvents()
 
     # -- Case 1: paused before the skip -> stays paused after
-    assert win.player.mpv.pause is True
-    before_pos = win.player.get_position()
-    win._seek_relative(5.0)
+    assert win.source_panel.player.mpv.pause is True
+    before_pos = win.source_panel.player.get_position()
+    win.source_panel.seek_relative(5.0)
     qapp.processEvents()
     time.sleep(0.2)
     qapp.processEvents()
-    after_pos = win.player.get_position()
+    after_pos = win.source_panel.player.get_position()
     print(f"paused case: before={before_pos:.3f}s after={after_pos:.3f}s")
-    assert win.player.mpv.pause is True, "should still be paused after skip"
-    assert win.play_button.text() == "Play"
+    assert win.source_panel.player.mpv.pause is True, "should still be paused after skip"
+    assert win.source_panel.play_button.text() == "Play"
     assert after_pos > before_pos + 4.0, "should have actually skipped forward"
     print("PASS: stays paused across arrow-key skip when it started paused")
 
     # -- Case 2: playing before the skip -> stays playing after
-    win.player.seek(0.0)
-    win.player.play()
-    win.play_button.setText("Pause")
+    win.source_panel.player.seek(0.0)
+    win.source_panel.player.play()
+    win.source_panel.play_button.setText("Pause")
     qapp.processEvents()
     time.sleep(0.2)
-    before_pos = win.player.get_position()
-    win._seek_relative(5.0)
+    before_pos = win.source_panel.player.get_position()
+    win.source_panel.seek_relative(5.0)
     qapp.processEvents()
     time.sleep(0.2)
     qapp.processEvents()
-    assert win.player.mpv.pause is False, "should still be playing after skip"
-    assert win.play_button.text() == "Pause"
-    after_pos = win.player.get_position()
+    assert win.source_panel.player.mpv.pause is False, "should still be playing after skip"
+    assert win.source_panel.play_button.text() == "Pause"
+    after_pos = win.source_panel.player.get_position()
     print(f"playing case: before={before_pos:.3f}s after={after_pos:.3f}s (should have jumped ~5s and kept playing)")
     assert after_pos > before_pos + 4.0, "should have actually skipped forward"
 
-    win.player.shutdown()
+    win.source_panel.shutdown()
+    win.preview_panel.shutdown()
     print("\nPRESERVE PLAY STATE SMOKE TEST PASSED")
 
 

scripts/smoke_test_preview.py

@@ -0,0 +1,141 @@
+"""Verifies the sync preview feature: building a short offset-corrected
+clip and playing it in the Preview tab (Kdenlive-style project/clip
+monitor tabs) instead of a popup dialog, plus the Sync markers panel
+showing Mark Audio Event before Mark Video Event with a Detect All Claps
+button alongside them.
+"""
+import sys
+import time
+from pathlib import Path
+
+from PyQt6.QtWidgets import QApplication
+
+sys.path.insert(0, str(Path(__file__).resolve().parent.parent))
+import main as app_main
+
+ROOT = Path(__file__).resolve().parent.parent
+CLIP = str(ROOT / "test_media" / "sync_test_OUTOFSYNC_audio+350ms.mp4")
+
+
+def main():
+    qapp = QApplication(sys.argv)
+    win = app_main.MainWindow()
+    win.show()
+    qapp.processEvents()
+
+    # -- Marker panel order: Mark Audio Event before Mark Video Event, with
+    # Detect All Claps in the same panel.
+    marker_layout = win.mark_audio_button.parentWidget().layout()
+    widgets = [marker_layout.itemAt(i).widget() for i in range(marker_layout.count())]
+    widgets = [w for w in widgets if w is not None]
+    audio_idx = widgets.index(win.mark_audio_button)
+    video_idx = widgets.index(win.mark_video_button)
+    detect_idx = widgets.index(win.detect_all_button)
+    assert audio_idx < video_idx, "Mark Audio Event should come before Mark Video Event"
+    print(f"PASS: marker panel order is audio({audio_idx}) < video({video_idx}), detect button present at {detect_idx}")
+
+    # -- Two tabs: Source (main player) and Preview
+    assert win.video_tabs.count() == 2
+    assert win.video_tabs.tabText(0) == "Source"
+    assert win.video_tabs.tabText(1) == "Preview"
+    assert win.video_tabs.widget(0) is win.source_panel
+    assert win.video_tabs.widget(1) is win.preview_panel
+    print("PASS: video area has Source and Preview tabs")
+
+    win._load(CLIP)
+    qapp.processEvents()
+    time.sleep(0.2)
+    qapp.processEvents()
+    assert win.video_tabs.currentIndex() == 0, "loading a new source should land on the Source tab"
+
+    win.source_panel.player.seek(5.000)
+    qapp.processEvents()
+    win._mark_video()
+
+    win.source_panel.player.seek(5.350)
+    qapp.processEvents()
+    win._mark_audio()
+
+    assert win.preview_button.isEnabled()
+    win.preview_button.click()
+    qapp.processEvents()
+    assert not win.preview_button.isEnabled(), "preview button should disable while building"
+    pending_path = win._pending_preview_path
+    print(f"building preview clip at: {pending_path}")
+
+    deadline = time.time() + 20
+    while win.video_tabs.currentIndex() != 1 and time.time() < deadline:
+        qapp.processEvents()
+        time.sleep(0.05)
+
+    assert win.video_tabs.currentIndex() == 1, "finishing a preview should switch to the Preview tab"
+    assert win.preview_button.isEnabled()
+    assert win._preview_temp_path == pending_path
+    assert Path(win._preview_temp_path).exists()
+    print("PASS: preview build switched to the Preview tab automatically")
+
+    qapp.processEvents()
+    time.sleep(0.3)
+    qapp.processEvents()
+
+    duration = win.preview_panel.player.mpv.duration
+    print(f"preview clip duration: {duration:.3f}s (expected ~{win.PREVIEW_DURATION_SEC}s, clamped to file bounds)")
+    assert duration is not None and 0 < duration <= win.PREVIEW_DURATION_SEC + 0.5
+    assert win.preview_panel.player.mpv.pause is False, "preview should auto-play"
+    assert win.preview_panel.play_button.text() == "Pause"
+
+    # -- The Preview tab has its own functional timeline (not a static video
+    # + a single play button): its own seek slider, sized to the preview
+    # clip's own duration, that seeks the preview player independently of
+    # the Source tab's slider/player.
+    assert win.preview_panel.seek_slider is not win.source_panel.seek_slider
+    preview_duration_ms = win.preview_panel.seek_slider.maximum()
+    assert preview_duration_ms > 0
+    print(f"Preview tab's own seek slider range: 0..{preview_duration_ms}ms")
+    target_ms = int(preview_duration_ms * 0.7)
+    win.preview_panel.seek_slider.setValue(target_ms)
+    win.preview_panel.seek_slider.sliderMoved.emit(target_ms)
+    qapp.processEvents()
+    time.sleep(0.2)
+    qapp.processEvents()
+    preview_pos = win.preview_panel.player.get_position()
+    source_pos = win.source_panel.player.get_position()
+    print(f"dragged Preview tab's own slider to 70%: preview player now at {preview_pos:.3f}s, source player still at {source_pos:.3f}s")
+    assert abs(preview_pos - target_ms / 1000.0) < 0.5, "Preview tab's slider should seek the preview player"
+    print("PASS: Preview tab has its own independent, functional timeline")
+
+    # -- Switching back to Source shouldn't disturb the main player, and the
+    # preview player keeps running independently (two live mpv instances).
+    main_pos_before_switch = win.source_panel.player.get_position()
+    win.video_tabs.setCurrentIndex(0)
+    qapp.processEvents()
+    main_pos_after_switch = win.source_panel.player.get_position()
+    print(f"main player position: before switch={main_pos_before_switch:.3f}s after={main_pos_after_switch:.3f}s")
+    assert abs(main_pos_after_switch - main_pos_before_switch) < 0.1, "switching tabs shouldn't move the main player"
+    assert win.source_panel.player.mpv.pause is True, "main player should still be paused (untouched by preview building)"
+    print("PASS: switching tabs doesn't disturb the main player")
+
+    # -- Building a second preview swaps the file without leaking the first
+    # temp file.
+    first_preview_path = win._preview_temp_path
+    win.preview_button.click()
+    qapp.processEvents()
+    deadline = time.time() + 20
+    while win._pending_preview_path is not None and time.time() < deadline:
+        qapp.processEvents()
+        time.sleep(0.05)
+    time.sleep(0.2)
+    qapp.processEvents()
+    assert not Path(first_preview_path).exists(), "first preview temp file should be cleaned up after being replaced"
+    assert Path(win._preview_temp_path).exists()
+    print("PASS: rebuilding the preview cleans up the previous temp file")
+
+    win.close()
+    qapp.processEvents()
+    assert not Path(win._preview_temp_path).exists() if win._preview_temp_path else True
+
+    print("\nPREVIEW TAB SMOKE TEST PASSED")
+
+
+if __name__ == "__main__":
+    main()

scripts/smoke_test_rightclick_filemenu.py

@@ -42,24 +42,25 @@ def main():
     print("PASS: Load File lives in the File menu")
 
     # -- Right-click on the player toggles play/pause
-    assert win.player.mpv.pause is True, "should start paused"
-    win.player.mpv.command("keypress", "MBTN_RIGHT")
+    assert win.source_panel.player.mpv.pause is True, "should start paused"
+    win.source_panel.player.mpv.command("keypress", "MBTN_RIGHT")
     qapp.processEvents()
     time.sleep(0.2)
     qapp.processEvents()
-    assert win.player.mpv.pause is False, "right-click should have started playback"
-    assert win.play_button.text() == "Pause"
+    assert win.source_panel.player.mpv.pause is False, "right-click should have started playback"
+    assert win.source_panel.play_button.text() == "Pause"
     print("PASS: right-click on player started playback")
 
-    win.player.mpv.command("keypress", "MBTN_RIGHT")
+    win.source_panel.player.mpv.command("keypress", "MBTN_RIGHT")
     qapp.processEvents()
     time.sleep(0.2)
     qapp.processEvents()
-    assert win.player.mpv.pause is True, "second right-click should pause again"
-    assert win.play_button.text() == "Play"
+    assert win.source_panel.player.mpv.pause is True, "second right-click should pause again"
+    assert win.source_panel.play_button.text() == "Play"
     print("PASS: right-click on player paused again")
 
-    win.player.shutdown()
+    win.source_panel.shutdown()
+    win.preview_panel.shutdown()
     print("\nRIGHT-CLICK / FILE MENU SMOKE TEST PASSED")
 
 

scripts/smoke_test_stream_ends_estimate.py

@@ -0,0 +1,157 @@
+"""Verifies the stream-ends offset estimate: it lands on the known offset of
+each synthetic fixture, it feeds Preview/Export exactly like a
+marker-derived offset does (including with no markers set at all), and
+Use Marker Offset brings the marker-derived value back afterwards without
+re-marking -- while the manual marker workflow itself behaves as before.
+
+Also covers the real-world shape this feature exists for, which none of
+the fixtures in test_media have: both tracks start at t=0 but the audio
+ends early (it started capturing late, and the recorder stopped both
+together). Generated on the fly as a 60s clip, long enough that the probe
+only reads the tail window instead of the whole file.
+"""
+import subprocess
+import sys
+import tempfile
+import time
+from pathlib import Path
+
+from PyQt6.QtWidgets import QApplication
+
+sys.path.insert(0, str(Path(__file__).resolve().parent.parent))
+import main as app_main
+from stream_edges import probe_stream_edges, TAIL_WINDOW_SEC
+
+ROOT = Path(__file__).resolve().parent.parent
+MEDIA = ROOT / "test_media"
+
+
+def wait_for(predicate, qapp, timeout=20):
+    deadline = time.time() + timeout
+    while not predicate() and time.time() < deadline:
+        qapp.processEvents()
+        time.sleep(0.05)
+    return predicate()
+
+
+def load(win, qapp, path):
+    win._load(str(path))
+    qapp.processEvents()
+    time.sleep(0.2)
+    qapp.processEvents()
+
+
+def estimate(win, qapp):
+    win.estimate_button.click()
+    qapp.processEvents()
+    assert wait_for(lambda: win.estimate_button.isEnabled(), qapp), "estimate did not finish in time"
+
+
+def make_audio_ends_early_clip(path: Path):
+    subprocess.run([
+        "ffmpeg", "-y", "-v", "error",
+        "-f", "lavfi", "-i", "testsrc=size=320x240:rate=30:duration=60",
+        "-f", "lavfi", "-i", "sine=frequency=1000:duration=59",
+        "-c:v", "libx264", "-preset", "ultrafast", "-c:a", "aac",
+        str(path),
+    ], check=True)
+
+
+def main():
+    # -- Probe alone, against every fixture's known offset.
+    expected = {
+        "sync_test_INSYNC.mp4": 0,
+        "sync_test_OUTOFSYNC_audio+350ms.mp4": -350,
+        "sync_test_OUTOFSYNC_video+350ms.mp4": 350,
+    }
+    for name, want in expected.items():
+        edges = probe_stream_edges(str(MEDIA / name))
+        print(f"{name}: end-based {edges.end_delay_ms:+.1f} ms, start-based {edges.start_delay_ms:+.1f} ms")
+        assert abs(edges.end_delay_ms - want) < 5, f"{name}: expected ~{want} ms"
+    print("PASS: end-based estimate matches every fixture's known offset")
+
+    with tempfile.TemporaryDirectory() as tmp:
+        clip = Path(tmp) / "audio_ends_early.mp4"
+        make_audio_ends_early_clip(clip)
+        assert 60 > TAIL_WINDOW_SEC, "clip should be long enough to exercise the tail-window read"
+        edges = probe_stream_edges(str(clip))
+        print(f"audio-ends-early clip: {edges}, end-based {edges.end_delay_ms:+.1f} ms")
+        assert abs(edges.end_delay_ms - 1000) < 30, "expected ~+1000 ms (delay audio)"
+        assert edges.estimates_disagree, "starts both at 0 but ends 1s apart -- should flag disagreement"
+    print("PASS: audio-ends-early clip estimates ~+1000 ms and flags the start/end disagreement")
+
+    qapp = QApplication(sys.argv)
+    win = app_main.MainWindow()
+    win.show()
+    qapp.processEvents()
+
+    assert not win.estimate_button.isEnabled(), "estimate needs a loaded file"
+    assert not win.use_marker_offset_button.isEnabled()
+
+    # -- Estimate with no markers: enables Preview/Export on its own.
+    load(win, qapp, MEDIA / "sync_test_OUTOFSYNC_video+350ms.mp4")
+    assert win.estimate_button.isEnabled() and win.estimate_action.isEnabled()
+    assert not win.export_button.isEnabled() and not win.preview_button.isEnabled()
+    estimate(win, qapp)
+    print(f"offset label after estimate: {win.offset_label.text()!r}")
+    assert win.offset.delay_ms == 350
+    assert win.override_spinbox.value() == 350 and win.override_spinbox.isEnabled()
+    assert win.export_button.isEnabled() and win.preview_button.isEnabled()
+    assert "estimated from stream ends" in win.offset_label.text()
+    assert not win.use_marker_offset_button.isEnabled(), "no markers yet, nothing to go back to"
+    assert win.video_mark is None and win.audio_mark is None, "estimate must not touch the markers"
+    print("PASS: estimate with no markers sets the offset and enables Preview/Export")
+
+    # -- Preview works from an estimate with no markers (centered on the playhead).
+    win.source_panel.player.seek(5.0)
+    qapp.processEvents()
+    win.preview_button.click()
+    qapp.processEvents()
+    assert wait_for(lambda: win.video_tabs.currentIndex() == 1, qapp), "preview did not build"
+    print("PASS: Preview builds from an estimated offset with no markers set")
+    win.video_tabs.setCurrentIndex(0)
+
+    # -- Manual markers behave as before and override the estimate.
+    load(win, qapp, MEDIA / "sync_test_OUTOFSYNC_audio+350ms.mp4")
+    assert win.offset_label.text() == "Offset: — (mark both events)"
+    win.source_panel.player.seek(5.000)
+    qapp.processEvents()
+    win._mark_video()
+    win.source_panel.player.seek(5.350)
+    qapp.processEvents()
+    win._mark_audio()
+    marker_label = win.offset_label.text()
+    print(f"offset label from markers: {marker_label!r}")
+    assert abs(win.offset.delay_ms - (-350)) < 40
+    assert marker_label == win.offset.describe()
+    assert win.use_marker_offset_button.isEnabled()
+    marker_delay = win.offset.delay_ms
+    print("PASS: manual markers set the offset exactly as before")
+
+    # -- Estimate replaces it; Use Marker Offset restores it without re-marking.
+    estimate(win, qapp)
+    assert win.offset.delay_ms == -349, f"got {win.offset.delay_ms}"
+    assert win.override_spinbox.value() == -349
+    assert win.video_mark is not None and win.audio_mark is not None, "estimate must not clear markers"
+    win.use_marker_offset_button.click()
+    qapp.processEvents()
+    assert win.offset.delay_ms == marker_delay
+    assert win.override_spinbox.value() == round(marker_delay)
+    assert win.offset_label.text() == marker_label
+    print("PASS: Use Marker Offset restores the marker-derived offset after an estimate")
+
+    # -- Loading a new file resets everything, including Use Marker Offset.
+    load(win, qapp, MEDIA / "sync_test_INSYNC.mp4")
+    assert not win.use_marker_offset_button.isEnabled()
+    assert not win.export_button.isEnabled()
+    estimate(win, qapp)
+    assert win.offset.delay_ms == 0
+    print("PASS: new file resets marker state; in-sync file estimates 0 ms")
+
+    win.close()
+    qapp.processEvents()
+    print("\nSTREAM ENDS ESTIMATE SMOKE TEST PASSED")
+
+
+if __name__ == "__main__":
+    main()

scripts/smoke_test_ui.py

@@ -30,14 +30,14 @@ def main():
     qapp.processEvents()
     time.sleep(0.3)
     qapp.processEvents()
-    print(f"loaded: duration slider range = {win.seek_slider.maximum()}ms")
+    print(f"loaded: duration slider range = {win.source_panel.seek_slider.maximum()}ms")
 
-    win.player.seek(5.000)
+    win.source_panel.player.seek(5.000)
     qapp.processEvents()
     win._mark_video()
     print(f"video mark: {win.video_mark:.3f}s -> label: {win.video_mark_label.text()}")
 
-    win.player.seek(5.350)
+    win.source_panel.player.seek(5.350)
     qapp.processEvents()
     win._mark_audio()
     print(f"audio mark: {win.audio_mark:.3f}s -> label: {win.audio_mark_label.text()}")
@@ -85,7 +85,8 @@ def main():
     assert Path(EXPORT_OUT).exists()
     print(f"exported file exists: {EXPORT_OUT}, size={Path(EXPORT_OUT).stat().st_size} bytes")
 
-    win.player.shutdown()
+    win.source_panel.shutdown()
+    win.preview_panel.shutdown()
     print("\nUI SMOKE TEST PASSED")
 
 

scripts/smoke_test_view_menu.py

@@ -31,7 +31,8 @@ def main():
     qapp.processEvents()
     assert not win.log_view.isVisible()
 
-    win.player.shutdown()
+    win.source_panel.shutdown()
+    win.preview_panel.shutdown()
     print("PASS: View > Show ffmpeg Log toggles the log panel correctly")
 
 

scripts/test_export_roundtrip.py

@@ -54,11 +54,11 @@ def find_audio_event_time(path: Path) -> float:
     return times[0]
 
 
-def run_case(fixture_name: str, expected_target: str, output_name: str):
+def run_case(fixture_name: str, expected_target: str, output_name: str, clip_start=None, clip_duration=None):
     input_path = ROOT / "test_media" / fixture_name
     output_path = OUT_DIR / output_name
 
-    print(f"Input fixture: {fixture_name}")
+    print(f"Input fixture: {fixture_name}" + (f" (clip {clip_start}+{clip_duration}s)" if clip_start else ""))
     video_mark = find_video_event_time(input_path)
     audio_mark = find_audio_event_time(input_path)
     print(f"  detected flash at t={video_mark:.3f}s, beep at t={audio_mark:.3f}s")
@@ -67,7 +67,7 @@ def run_case(fixture_name: str, expected_target: str, output_name: str):
     print(f"  {offset.describe()}")
     assert offset.target == expected_target, f"expected {expected_target} branch, got {offset.target}"
 
-    result = export_synced(str(input_path), offset.delay_ms, str(output_path))
+    result = export_synced(str(input_path), offset.delay_ms, str(output_path), clip_start=clip_start, clip_duration=clip_duration)
     print(f"  ran: {' '.join(result.command)}")
 
     after_video_t = find_video_event_time(output_path)
@@ -84,7 +84,17 @@ def main():
     OUT_DIR.mkdir(exist_ok=True)
     run_case("sync_test_OUTOFSYNC_audio+350ms.mp4", "video", "_export_roundtrip_negative.mp4")
     run_case("sync_test_OUTOFSYNC_video+350ms.mp4", "audio", "_export_roundtrip_positive.mp4")
-    print("PASS: both ffmpeg command branches (positive and negative delay) round-trip correctly.")
+    # Clip mode (used by the GUI's preview feature) is a genuinely different
+    # code path -- ffdelay's own documented quirk is that -ss/-t must be
+    # applied to BOTH inputs, in the right position relative to -itsoffset,
+    # or the clip covers different real-world windows per input and the
+    # "preview" would silently show wrong sync. Exercise it explicitly
+    # rather than assuming the full-file case generalizes.
+    run_case(
+        "sync_test_OUTOFSYNC_audio+350ms.mp4", "video", "_export_roundtrip_clip_mode.mp4",
+        clip_start="2.000", clip_duration=8,
+    )
+    print("PASS: both ffmpeg command branches (positive and negative delay) round-trip correctly, including clip mode.")
 
 
 if __name__ == "__main__":

stream_edges.py

@@ -0,0 +1,145 @@
+"""Offset estimation from where the video and audio tracks start and end,
+as an optional alternative to marking a sync event by hand.
+
+Premise: a recorder that stops both tracks at the same real-world moment,
+but whose audio started capturing late (or early) while still being
+written as starting at t=0, produces a file where the audio runs out
+before (or after) the video by exactly the sync error. So
+
+    delay_ms = (video_end - audio_end) * 1000
+
+using the same sign convention as sync_logic.compute_offset (positive:
+audio is ahead and gets delayed; negative: video gets delayed).
+
+This is only an *estimate*: a file whose audio genuinely stopped early
+(a mic cutting out before the camera did) looks identical but has nothing
+to do with sync. The start-based figure is reported alongside it as a
+cross-check -- it's what carries the offset when the container encodes it
+as a stream start_time instead, and a disagreement between the two is
+worth surfacing.
+"""
+import subprocess
+from dataclasses import dataclass
+from typing import List, Optional, Tuple
+
+# How far before the end of the file to start reading packets. ffprobe
+# seeks there instead of scanning the whole file, which keeps this fast on
+# multi-hour recordings. If a track ends earlier than this window (so no
+# packets for it turn up), the whole track is scanned instead.
+TAIL_WINDOW_SEC = 30.0
+
+# Start- and end-based estimates are compared against this. AAC encoder
+# delay alone typically puts the start ~20-45ms off, so anything much
+# tighter would warn on perfectly ordinary files.
+DISAGREEMENT_TOLERANCE_MS = 50.0
+
+
+class StreamEdgesError(RuntimeError):
+    pass
+
+
+@dataclass(frozen=True)
+class StreamEdges:
+    video_start: float
+    video_end: float
+    audio_start: float
+    audio_end: float
+
+    @property
+    def end_delay_ms(self) -> float:
+        return (self.video_end - self.audio_end) * 1000.0
+
+    @property
+    def start_delay_ms(self) -> float:
+        return (self.video_start - self.audio_start) * 1000.0
+
+    @property
+    def estimates_disagree(self) -> bool:
+        return abs(self.end_delay_ms - self.start_delay_ms) > DISAGREEMENT_TOLERANCE_MS
+
+
+def _ffprobe(args: List[str]) -> str:
+    result = subprocess.run(["ffprobe", "-v", "error", *args], capture_output=True, text=True)
+    if result.returncode != 0:
+        raise StreamEdgesError(f"ffprobe failed: {result.stderr.strip()}")
+    return result.stdout
+
+
+def _parse_float(value: str) -> Optional[float]:
+    try:
+        return float(value)
+    except ValueError:
+        return None  # "N/A"
+
+
+def _format_duration(input_path: str) -> Optional[float]:
+    out = _ffprobe(["-show_entries", "format=duration", "-of", "csv=p=0", input_path])
+    return _parse_float(out.strip())
+
+
+def _stream_start(input_path: str, stream_specifier: str) -> float:
+    out = _ffprobe([
+        "-select_streams", stream_specifier,
+        "-show_entries", "stream=start_time", "-of", "csv=p=0", input_path,
+    ])
+    start = _parse_float(out.strip())
+    return start if start is not None else 0.0
+
+
+def _packets(input_path: str, stream_specifier: str, read_from: Optional[float]) -> List[Tuple[float, Optional[float]]]:
+    args = ["-select_streams", stream_specifier]
+    if read_from is not None:
+        args += ["-read_intervals", f"{read_from:.3f}%"]
+    args += ["-show_entries", "packet=pts_time,duration_time", "-of", "csv=p=0", input_path]
+    packets = []
+    for line in _ffprobe(args).splitlines():
+        fields = line.strip().rstrip(",").split(",")
+        pts = _parse_float(fields[0]) if fields and fields[0] else None
+        if pts is None:
+            continue
+        duration = _parse_float(fields[1]) if len(fields) > 1 else None
+        packets.append((pts, duration))
+    return packets
+
+
+def _stream_end(input_path: str, stream_specifier: str, file_duration: Optional[float]) -> float:
+    """End of the last packet (pts + duration), not the container's stored
+    duration -- that's frequently missing (e.g. per-stream in MKV) or
+    rounded, and this needs to be accurate to well under a frame."""
+    read_from = None
+    if file_duration is not None and file_duration > TAIL_WINDOW_SEC:
+        read_from = file_duration - TAIL_WINDOW_SEC
+    packets = _packets(input_path, stream_specifier, read_from)
+    if not packets and read_from is not None:
+        packets = _packets(input_path, stream_specifier, None)
+    if not packets:
+        raise StreamEdgesError(f"no packets found for stream {stream_specifier}")
+
+    # Max rather than last: video with B-frames isn't stored in pts order.
+    ends = [pts + duration for pts, duration in packets if duration is not None]
+    if ends:
+        return max(ends)
+    # No per-packet durations (some containers): fall back to the gap
+    # between the last two packets as the final packet's duration.
+    pts_sorted = sorted(pts for pts, _ in packets)
+    step = pts_sorted[-1] - pts_sorted[-2] if len(pts_sorted) > 1 else 0.0
+    return pts_sorted[-1] + step
+
+
+def probe_stream_edges(
+    input_path: str,
+    video_stream_index: Optional[int] = None,
+    audio_stream_index: Optional[int] = None,
+) -> StreamEdges:
+    """Stream indices are ffmpeg global indices (mpv's `ff-index`), so the
+    estimate uses the same tracks the player is showing -- see
+    clap_detection.py for why that matters on multi-track files."""
+    video_spec = f"{video_stream_index}" if video_stream_index is not None else "v:0"
+    audio_spec = f"{audio_stream_index}" if audio_stream_index is not None else "a:0"
+    file_duration = _format_duration(input_path)
+    return StreamEdges(
+        video_start=_stream_start(input_path, video_spec),
+        video_end=_stream_end(input_path, video_spec, file_duration),
+        audio_start=_stream_start(input_path, audio_spec),
+        audio_end=_stream_end(input_path, audio_spec, file_duration),
+    )

test_media/README.md

@@ -31,3 +31,10 @@ validating marker capture, offset math, and lossless export.
 
 Frame at exactly t=5.000s is frame index 150 (30fps) — useful for checking
 frame-step precision lands exactly on it.
+
+- `sync_test_MULTITRACK.mp4` — 10s, blue video, two AAC audio tracks: track
+  index 1 (default) has a beep at t=3.000s, track index 2 (not default) has
+  a beep at t=7.000s. Used to verify clap detection follows whichever audio
+  track mpv is actually playing (its `ff-index`), not whichever one
+  ffmpeg's own default-stream selection would pick -- see
+  scripts/smoke_test_multitrack_clap_detection.py.