ElevenLabs
ElevenLabs Scribe provides high-accuracy transcription across many languages with keyterm prompting, diarization, non-speech event tagging, and entity options.
API documentation: ElevenLabs speech-to-text
Models
| Model ID | Notes |
|---|---|
elevenlabs/scribe_v2 | Latest Scribe model. Supports v2-only options such as no_verbatim, entity options, and speaker role detection. |
elevenlabs/scribe_v1 | Previous Scribe model. |
Params
toml
[elevenlabs.scribe_v2.params]
language_code = "sv"
diarize = true
keyterms = ["OSTT", "Scribe", "Whisper"]
tag_audio_events = true
timestamps_granularity = "word"bash
ostt transcribe interview.mp3 -m elevenlabs/scribe_v2 --param diarize=true --param keyterms=OSTT,Scribe
ostt model params elevenlabs/scribe_v2 --format json| Param | Type | Description |
|---|---|---|
language_code | string | Optional language code. Set when known to improve accuracy. |
keyterms | string list | Domain vocabulary. Saved ostt keyword terms are used as fallback only when keyterms is not set. |
tag_audio_events | boolean | Tag non-speech events. |
timestamps_granularity | string | Timestamp granularity: none, word, or character. |
diarize | boolean | Enable speaker diarization. |
num_speakers | integer | Expected number of speakers, from 1 to 32. |
diarization_threshold | number | Speaker diarization threshold from 0.1 to 0.4. Requires diarize = true and cannot be combined with num_speakers. |
temperature | number | Sampling temperature. ElevenLabs accepts 0.0 to 2.0. |
file_format | string | Input audio format: other or pcm_s16le_16. |
seed | integer | Deterministic seed. |
use_multi_channel | boolean | Use multi-channel input handling. |
detect_speaker_roles | boolean | Scribe v2 only. Detect speaker roles. Requires diarize = true and cannot be combined with use_multi_channel = true. |
no_verbatim | boolean | Scribe v2 only. Reduce verbatim artifacts. |
entity_detection | string or string list | Scribe v2 only. Entity detection setting, such as all, pii, phi, pci, other, or offensive_language. |
entity_redaction | string or string list | Scribe v2 only. Entity redaction setting. Must be a subset of entity_detection. |
entity_redaction_mode | string | Scribe v2 only. Redaction mode: redacted, entity_type, or enumerated_entity_type. |