Home Index Transcribe

Home Index Transcribe is an RPC module for Home Index. It uses WhisperX to generate transcripts and subtitle files from audio and video. The module implements the RPC interface described in the Home Index project so that transcripts can be indexed and searched.

Quick start

The docker-compose.yml in this repository launches a small stack containing Home Index, Meilisearch and this transcribe module. After installing Docker, run:

docker compose up

Drop media files into bind-mounts/files/ and Home Index will automatically invoke the module to produce transcripts. Metadata and cache files are stored under bind-mounts/.

Configuration

Several environment variables control how WhisperX runs:

• NAME – module name (transcribe by default)
• DEVICE – compute device (cuda or cpu)
• WHISPER_MODEL – Whisper model variant (e.g. medium)
• BATCH_SIZE – batch size for transcription
• COMPUTE_TYPE – computation precision (int8 by default)
• LANGUAGE – language code (en by default)
• PYANNOTE_DIARIZATION_AUTH_TOKEN – token for speaker diarization
• BASE_TIMESTAMP_PATH – dot-delimited path within the document containing the creation timestamp used when prefixing transcript chunks

Running manually

If you want to start the RPC server without Docker:

pip install -r requirements.txt
python packages/home_index_transcribe/main.py

The server listens on 0.0.0.0:9000. Add its endpoint to the MODULES environment variable of Home Index so it is called during indexing.

More information

See the Home Index documentation for details on the RPC module specification and additional configuration options.