Voice (STT/TTS)

abTARS transcribes voice messages and can respond with synthesized speech.

Speech-to-Text (STT)

Provider: Groq Whisper (whisper-large-v3-turbo).

Setup

# ~/.abtars/config/.env
STT_ENABLED=true
GROQ_API_KEY=<secret>

Flow

1. User sends a voice note on Telegram
2. Bridge downloads the audio file
3. Sends to Groq Whisper API for transcription
4. Injects transcript as: [🎤 voice, EN] Hello, how are you?
5. Agent responds normally (text or TTS)

Language handling

• LANGUAGE_HINT_PROMPT guides Whisper (e.g. "ez egy magyar szöveg. or English")
• Whisper returns detected language code (hu, en, ja, etc.)
• users.json defines expected languages per user: "languages": ["hu", "en"]
• If detected language isn't in the user's list → agent asks back (likely hallucination on short audio)
• Soft check via prompt — no hard rejection

Text-to-Speech (TTS)

When enabled, the agent can respond with voice messages on Telegram.

Setup

TTS_ENABLED=true
TTS_PROVIDER=openai    # or other supported provider

Platform support

Platform	STT	TTS
Telegram	✓	✓
Discord	—	—
IRC	—	—

Voice is Telegram-only — it's the only platform that sends voice note file IDs the bridge can download.