How welcomyou/sherpa-vietnamese-asr works
Sherpa Vietnamese ASR is an offline-first Vietnamese transcription product with desktop and web-service interfaces. The main workflow is audio submission, decoding and ASR, followed by optional punctuation, speaker diarization, overlap separation, and quality analysis; users can review and edit results. The web service adds authenticated sessions, queued jobs, persistence, and meeting/search features. README documents an offline PWA, but the sampled browser pipeline explicitly says inference is not implemented in the baseline scaffold. Some subsystem links below are documented capabilities rather than inspected call paths.
Main entry points: asr_engine.py ASR engine · server.py Web API · app.py Desktop app · queue_manager.py Job queue
Components 26
Desktop experience
- Desktop app
app.py - File transcription
tab_file.py - Live recording
tab_live.py
Recognition pipeline
- Transcriber pipeline
transcriber.py - ASR engine
core/asr_engine.py - Audio decoding
core/audio_decode.py - Voice activity detection
core/vad_utils.py - Punctuation restoration
core/punctuation_restorer_improved.py - Speaker diarization
core/speaker_diarization_pure_ort.py - CAM++ diarization
core/speaker_diarization_senko_campp.py - Overlap separation
core/overlap_separator.py - Audio quality
core/audio_analyzer.py
Web service
- Web API
web_service/server.py - Authentication
web_service/auth.py - Session lifecycle
web_service/session_manager.py - Job queue
web_service/queue_manager.py - Service database
web_service/database.py - Meeting summarizer
web_service/summarizer.py
Offline PWA
- PWA file server
offline_pwa/server.py - Browser pipeline scaffold
offline_pwa/src/engine/pipeline.ts
Runtime and acceleration
- Model and app config
core/config.py - GPU calibration
core/calibration.py - Hardware providers
core/hardware_accel.py - Service config
web_service/config.py
Other components
- Desktop user
- Web user
Connections 36
- Desktop user → Desktop app: opens
README.md - Desktop app → File transcription: offers file workflow
app.py - Desktop app → Live recording: offers recording
app.py - File transcription → ASR engine: invokes
tab_file.py - File transcription → Audio decoding: uses
tab_file.py - Live recording → Audio quality: analyzes audio
tab_live.py - ASR engine → Audio decoding: decodes through
core/asr_engine.py - ASR engine → Voice activity detection: uses
core/asr_engine.py - ASR engine → Punctuation restoration: uses
core/asr_engine.py - ASR engine → Speaker diarization: uses
core/asr_engine.py - ASR engine → CAM++ diarization: uses
core/asr_engine.py - ASR engine → Overlap separation: uses
core/asr_engine.py - Desktop user → File transcription: submits audio
README.md - Web user → Web API: uploads audio
README.md - Web API → Authentication: checks access
web_service/server.py - Web API → Session lifecycle: tracks sessions
web_service/server.py - Web API → Job queue: enqueues jobs
web_service/server.py - Job queue → Service database: updates jobs
web_service/queue_manager.py - Job queue → ASR engine: runs transcription
web_service/queue_manager.py - Job queue → Meeting summarizer: dispatches summaries
web_service/queue_manager.py - Web API → Service database: reads and writes
web_service/server.py - Session lifecycle → Service database: stores session state
web_service/session_manager.py - Web API → GPU calibration: requests calibration
web_service/server.py - GPU calibration → Hardware providers: detects providers
core/calibration.py - GPU calibration → ASR engine: benchmarks stages
core/calibration.py - Desktop app → Model and app config: loads settings
app.py - ASR engine → Model and app config: reads model settings
core/asr_engine.py - Web API → Service config: uses settings
web_service/server.py - Service config → Model and app config: shares model config
web_service/config.py - Web user → PWA file server: opens offline app
README.md - PWA file server → Browser pipeline scaffold: serves browser shell
- Browser pipeline scaffold → ASR engine: inference not implemented
- ASR engine → Audio quality: evaluates quality
README.md - ASR engine → Punctuation restoration: restores punctuation
README.md - ASR engine → Speaker diarization: assigns speakers
README.md - ASR engine → Overlap separation: separates overlap
README.md