How oldgeorsk/canto-transcription works
This repository is a Cantonese interview transcription tool with an optional, independent speech-detection step. Its principal workflow takes interview audio through WhisperX transcription, optionally adds speaker diarization, and produces a timestamped raw transcript; researchers can then use an external AI platform and the supplied guide to create normalized written Chinese. VAD creates speech reports and optional clips but does not feed the transcription script. The README documents several integrations and output behaviors; sampled source excerpts show model use and parameter handling, but not every orchestration detail.
Main entry points: transcribe.py Transcription Runner · vad.py VAD Runner · vad_config.json VAD Settings · transcribe_config_whisperx.json WhisperX Settings
Components 16
Optional VAD
- VAD Runner
vad.py - VAD Settings
vad_config.json - Silero VAD
- Speech Report
- Speech Clips
Raw Transcription
- Transcription Runner
transcribe.py - WhisperX Settings
transcribe_config_whisperx.json - WhisperX ASR
- Speaker Diarization
- HuggingFace Token
- Raw Transcript
Text Normalization
- Cantonese Writing Guide
skills/spoken-to-written-Canto.md - Normalized Text
Other components
- Researcher
- Interview Audio
- AI Platform
Connections 18
- Researcher → Interview Audio: provides
README.md - Researcher → VAD Runner: runs optionally
README.md - Interview Audio → VAD Runner: analyzes
README.md - VAD Settings → VAD Runner: configures
README.md - VAD Runner → Silero VAD: detects speech
vad.py - VAD Runner → Speech Report: writes report
README.md - VAD Runner → Speech Clips: writes clips
README.md - Researcher → Transcription Runner: runs
README.md - Interview Audio → Transcription Runner: transcribes
README.md - WhisperX Settings → Transcription Runner: configures
README.md - Transcription Runner → WhisperX ASR: runs ASR
transcribe.py - Transcription Runner → Speaker Diarization: diarizes optionally
README.md - HuggingFace Token → Transcription Runner: authorizes
README.md - Transcription Runner → Raw Transcript: writes transcript
README.md - Researcher → AI Platform: requests normalization
README.md - Raw Transcript → AI Platform: provides text
README.md - Cantonese Writing Guide → AI Platform: guides conversion
README.md - AI Platform → Normalized Text: produces text
README.md