ElevenLabs

Scribe — high-accuracy STT from the voice company.

All prices are the vendor’s public list price — 0% markup. Extra provider knobs pass through untouched via provider_params.

Model	Modes	List price	Diarization	Word timings
elevenlabs/scribe_v2_realtime	streaming	$0.3888/hr	—	✓
elevenlabs/scribe_v2	batch	$0.21996/hr	✓	✓
elevenlabs/scribe_v1	batch	$0.21996/hr	✓	✓

Provider options

Reach past the unified surface with provider_params — forwarded streaming → session fields; batch → form fields. Typed in the SDKs as {provider}Params interfaces (providerParams option / provider_params= kwarg).

Param	Type	Default	Applies to	What it does
commit_strategy	manual · vad	—	streaming	When transcripts commit: on silence (vad) or on demand
vad_threshold	number	—	streaming	Voice-activity sensitivity for vad commits
vad_silence_threshold_secs	number	—	streaming	Silence that triggers a vad commit
secondary_languages	array	—	streaming	Additional expected languages
no_verbatim	boolean	false	streaming	Clean up disfluencies
filter_background_audio	boolean	false	streaming	Suppress non-speech audio
enable_logging	boolean	true	streaming	false = zero-retention mode at ElevenLabs
num_speakers	integer	—	batch	Expected speaker count for diarization
timestamps_granularity	none · word · character	—	batch	Timestamp detail level
tag_audio_events	boolean	false	batch	Mark laughter, applause, and other audio events
entity_detection	boolean	false	batch	Detect 65 entity types (extra vendor charge)
detect_speaker_roles	boolean	false	batch	Label agent/customer roles

Try it

curl -s https://api.speechrouter.ai/v1/audio/transcriptions \
  -H "Authorization: Bearer $SPEECHROUTER_API_KEY" \
  -F model=elevenlabs/scribe_v2 \
  -F file=@audio.wav

_{Generated from the gateway catalog — the billing engine’s own source of truth.}