Skip to content

SPEV: turn decisions up to 36% sooner. Details →

Speech-to-text models

The models that turn audio into text, and which one to pick for transcription, translation, or conversation.

Two models take audio in and return text. Pass the id in the model field.

ModelModel IDEndpointUse it for
Rhythmrhythm/v1/audio/transcriptions, or a realtime transcription sessionWriting down the words that were said
Symphonysymphony/v1/audio/transcriptions, /v1/audio/translations, /v1/chat/completions, or a realtime sessionWriting down the words, translating them, or replying to them

Rhythm

A transcription model. It returns the transcript, with timestamps and diarization as structured fields.

FieldValue
Model IDrhythm
ModalitiesAudio in, text out
EndpointPOST /v1/audio/transcriptions, or a realtime transcription session
Parametersresponse_format, timestamp_granularities[], speaker_labels, num_speakers

REST has the request contract.

Symphony

A multimodal model that takes text, image, audio, and video in and returns text and speech. It transcribes on POST /v1/audio/transcriptions and translates on POST /v1/audio/translations or in a realtime translation session. It answers on POST /v1/chat/completions, and in a realtime session it hears the user and replies out loud.

Use Symphony when you want a reply rather than a transcript, or when one model should do both: it hears and answers in one step instead of transcribing and then prompting separately.

FieldValue
Model IDsymphony
ModalitiesText, image, audio, video in; text and audio out
EndpointPOST /v1/audio/transcriptions, POST /v1/audio/translations, POST /v1/chat/completions, or a realtime session

On the chat endpoint it takes the standard generation parameters, such as max_tokens, temperature, and response_format. The model catalog lists the full set. Speech translation has the translation request contract.

The model catalog carries the live list, and pricing has the current rates. Languages has language coverage.