Skip to content

AssemblyAI

Streaming turn model built for voice agents, plus strong async models.

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
assemblyai/universal-3-5-pro streaming $0.45/session-hr
assemblyai/universal-streaming-english streaming $0.15012/session-hr
assemblyai/universal-streaming-multilingual streaming $0.15012/session-hr
assemblyai/universal-2 batch $0.20988/hr

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

Param Type Default Applies to What it does
end_of_turn_confidence_threshold number 0.4 streaming Confidence needed to declare end of turn
min_turn_silence integer streaming Minimum silence (ms) before a turn can end
max_turn_silence integer streaming Silence (ms) that forces end of turn
vad_threshold number streaming Voice-activity detection sensitivity
continuous_partials boolean false streaming Emit partials continuously instead of on change
max_speakers integer streaming Cap for streaming diarization
language_detection boolean false streaming · batch Auto-detect the spoken language
domain medical-v1 streaming Domain-tuned recognition
prompt string streaming Free-text context to bias recognition
redact_pii boolean false streaming · batch Redact personally identifiable information
inactivity_timeout integer streaming Close the vendor session after this many silent seconds
mode max_accuracy · balanced · min_latency streaming universal-3-5-pro: latency/accuracy preset
Terminal window
curl -s https://api.speechrouter.ai/v1/audio/transcriptions \
-H "Authorization: Bearer $SPEECHROUTER_API_KEY" \
-F model=assemblyai/universal-2 \
-F file=@audio.wav

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