- Cache loaded audio in _sf_load() — previously the entire WAV file was
re-read from disk for every 10s crop call. For a 3-hour file with
1000+ chunks, this meant ~345GB of disk reads. Now read once, cached.
- Better progress messages for long files: show elapsed time in m:ss
format, warn "(180min audio, this may take a while)" for files >10min
- Increased progress poll interval from 2s to 5s (less noise)
- Better time estimate: use 0.8x audio duration (was 0.5x)
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Diarization: Audio.crop patch now pads short segments with zeros to
match the expected duration. pyannote batches embeddings with vstack
which requires uniform tensor sizes — the last segment of a file can
be shorter than the 10s window.
CI: Reordered sidecar workflow to check for python/ changes FIRST,
before bumping version or configuring git. All subsequent steps are
gated on has_changes. This prevents unnecessary version bumps and
build runs when only app code changes.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The previous patch only replaced Audio.__call__ (segmentation), but
pyannote also calls Audio.crop during speaker embedding extraction.
crop loads a time segment of audio — patched to load full file via
soundfile then slice the tensor to the requested time range.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
torchaudio 2.10 unconditionally delegates load() to torchcodec, ignoring
the backend parameter. Since torchcodec is excluded from PyInstaller,
this broke our pyannote Audio monkey-patch.
Fix: replace torchaudio.load() with soundfile.read() + torch.from_numpy().
soundfile handles WAV natively (audio is pre-converted to WAV), has no
torchcodec dependency, and is already a transitive dependency.
Also added soundfile to PyInstaller hiddenimports.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
CSP: Add blob: to connect-src/img-src/media-src for wavesurfer.js audio
playback. Add http://tauri.localhost to default-src for devtools.
pyannote: sys.modules block didn't work — pyannote still uses AudioDecoder
unconditionally. New approach: monkey-patch Audio.__call__ in diarize.py
to use torchaudio.load() directly, bypassing the broken torchcodec path.
Patch runs once before pipeline loading.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- transcribe: catch model load failures on CUDA and retry with CPU
- hardware detect: test CUDA runtime actually works (torch.zeros on cuda)
before recommending GPU, since CPU-only builds detect CUDA via driver
but lack cublas/cuDNN libraries
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Move the blocking pipeline() call to a daemon thread and emit estimated
progress messages every 2 seconds from the main thread. The progress
estimate uses audio duration to calibrate the expected total time.
Also pass audio_duration_sec from PipelineService to DiarizeService.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Split files >1 hour into 5-minute chunks via ffmpeg, transcribe each
chunk independently, then merge results with corrected timestamps.
Also add chunk-level progress markers every 10 segments for all files.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Previously, progress messages were only sent every 5th segment due to
a `segment_count % 5` guard. This made the UI feel unresponsive for
short recordings with few segments. Now every segment emits a progress
update with a more descriptive message including the segment number
and audio percentage.
Adds a test verifying that all 8 mock segments produce progress
messages, not just every 5th.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Send each segment to the frontend immediately after transcription via
a new pipeline.segment IPC message, then send speaker assignments as a
batch pipeline.speaker_update message after diarization completes. This
lets the UI display segments progressively instead of waiting for the
entire pipeline to finish.
Changes:
- Add partial_segment_message and speaker_update_message IPC factories
- Add on_segment callback parameter to TranscribeService.transcribe()
- Emit partial segments and speaker updates from PipelineService.run()
- Add send_and_receive_with_progress to SidecarManager (Rust)
- Route pipeline.segment/speaker_update events in run_pipeline command
- Listen for streaming events in Svelte frontend (+page.svelte)
- Add tests for new message types, callback signature, and update logic
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Convert non-WAV audio to 16kHz mono WAV before diarization (pyannote
v4.0.4 AudioDecoder returns None duration for FLAC, causing crash)
- Handle pyannote 4.0 DiarizeOutput return type (unwrap .speaker_diarization)
- Disable pyannote telemetry (np.isfinite(None) bug with max_speakers)
- Use huggingface_hub.login() to persist token for all sub-downloads
- Pre-download sub-models (segmentation-3.0, speaker-diarization-community-1)
- Add third required model license link in settings UI
- Improve SpeakerManager hints based on settings state
- Add word-wrap to transcript text
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Add "Speakers" tab in Settings with HF token input field
- Include step-by-step instructions for obtaining the token
- Pass hf_token from settings through Rust → Python pipeline → diarize
- Token can also be set via HF_TOKEN environment variable as fallback
- Move skip_diarization checkbox to Speakers tab
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Replace progress bar with task checklist showing pipeline steps
(load model, transcribe, load diarization, identify speakers, merge)
- Fix WaveformPlayer: track ready state, disable controls until loaded,
play from current position instead of resetting to start
- Fix workspace height calc to prevent bottom content cutoff
- Show HF_TOKEN setup hint in SpeakerManager when no speakers detected
- Add console logging for progress events to aid debugging
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Enhanced ProgressOverlay with spinner, better styling, and z-index 9999
- Import button shows "Processing..." with pulse animation while transcribing
- Fix pyannote API: use token= instead of deprecated use_auth_token=
- Read HF_TOKEN from environment for pyannote model download
- Add console logging for click-to-seek debugging
- Add color-scheme: dark for native form controls
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Stream pipeline progress to frontend via Tauri events so the progress
overlay updates in real time during transcription/diarization
- Gracefully fall back to transcription-only when diarization fails
(e.g. pyannote not installed) instead of erroring the whole pipeline
- Add color-scheme: dark to fix native select/option elements rendering
with unreadable white backgrounds
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Implement ExportService using pysubs2 for caption formats (SRT, VTT, ASS)
and custom formatters for plain text and Markdown
- SRT exports with [Speaker]: prefix, WebVTT with <v Speaker> voice tags,
ASS with color-coded speaker styles
- Plain text groups by speaker with labels, Markdown adds timestamps
- Add export.start IPC handler and export_transcript Tauri command
- Add export dropdown menu in header (appears after transcription)
- Uses native save dialog for output file selection
- Add pysubs2 dependency
- Tests: 30 Python (6 export tests), 6 Rust, 0 Svelte errors
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
