Matrix Voice Transcript

Transcription languages / Языки распознавания / Sprachen / Langues: 🇬🇧 English · 🇷🇺 Русский · 🇩🇪 Deutsch · 🇫🇷 Français · 🇪🇸 Español · 🇵🇹 Português · 🇵🇱 Polski · 🇨🇿 Czech · 🇸🇰 Slovak · 🇧🇬 Bulgarian · 🇺🇦 Ukrainian · 🇳🇱 Dutch · 🇮🇹 Italian · 🇷🇴 Romanian · 🇸🇪 Swedish · 🇩🇰 Danish · 🇫🇮 Finnish · 🇳🇴 Norwegian · 🇬🇷 Greek · 🇭🇺 Hungarian · 🇪🇪 Estonian · 🇱🇻 Latvian · 🇱🇹 Lithuanian · 🇸🇮 Slovenian · 🇭🇷 Croatian · 🇲🇹 Maltese

Matrix bot that transcribes voice messages and audio files using NVIDIA NeMo Parakeet TDT running locally on CPU. Supports E2EE rooms. No audio leaves the server.

Requirements

• Matrix bot account with an access token.
• MATRIX_PASSWORD recommended for E2EE rooms (enables stale-device pruning; without it decryption may fail on first run).
• ~2.5 GB disk space for the model checkpoint (cached in ./models, downloaded on first start).

Quick start

1. Copy .env.example to .env and fill in the variables.
2. docker compose up -d
3. Invite the bot to a Matrix room.

Environment variables

Variable	Description
MATRIX_HS_URL	Homeserver URL (with https://)
MATRIX_USER_ID	Full bot MXID, e.g. @voicebot:example.org
MATRIX_ACCESS_TOKEN	Bot access token
MATRIX_PASSWORD	Optional. Prunes stale E2EE devices on startup; required for reliable decryption in encrypted rooms.
LOCALE	Message language: en (default) or ru
ASR_MODEL_NAME	NeMo model (default: nvidia/parakeet-tdt-0.6b-v3)
MAX_AUDIO_BYTES	Max file size in bytes (default: 26214400 = 25 MB)
ASR_CHUNK_SECONDS	Splits long audio into chunks of this length before transcription, to bound peak memory (default: 60)
STORE_PATH	Olm key store path inside the container (default: /data/store)

Supported formats: ogg/opus, webm, mp4/m4a, aac, flac, mp3, wav.

Message language / Смена языка

LOCALE=en   # English (default)
LOCALE=ru   # Russian / Русский

docker compose restart to apply.

Local development

Requires Python 3.11+ and ffmpeg on PATH.

python -m venv .venv && .venv\Scripts\activate   # Windows
pip install -r requirements.txt
python -m src.main

Security

• Never commit .env.
• Transcribed text is never written to logs.
• Temp audio files are deleted immediately after transcription.

License

MIT