speech-to-text

Call client.speech_to_text.convert with the audio file and model_id set to scribe_v2. The response returns the full text, detected language, and optional word-level timestamps. Python, JavaScript, and cURL are all supported.

Set diarize=true in the convert request to label each word with a speaker_id, supporting up to 32 speakers. You can optionally set num_speakers or tune diarization_threshold for sensitivity control.

Supported audio formats include MP3, WAV, M4A, FLAC, OGG, WebM, AAC, AIFF, and Opus; video formats include MP4, AVI, MKV, MOV, WMV, FLV, WebM, MPEG, and 3GPP. Files can be up to 3GB and 10 hours long.

Yes, the scribe_v2_realtime model provides streaming transcription with roughly 150ms latency over WebSocket. It emits partial transcripts for live feedback and committed transcripts as final results, with manual or VAD-based commit strategies.

A 401 error means the API key is invalid or missing. Set the ELEVENLABS_API_KEY environment variable or pass the key directly when creating the client, and verify the key in your ElevenLabs dashboard.

Use the keyterms parameter to bias the model toward up to 100 specific terms such as brand names or technical vocabulary. Each term must be 50 characters or fewer and contain at most 5 words.