Skip to main content
Using a coding agent? Install the together-audio skill to let your agent write correct speech-to-text code automatically. See agent skills for details.
Together AI hosts speech recognition models including OpenAI’s Whisper and NVIDIA Parakeet for batch transcription and real-time streaming.
Read the end-to-end guide to build a live voice agent powered by Together AI’s real-time STT and TTS pipeline.

​
Quickstart

Basic transcription and translation: 
from pathlib import Path

from together import Together

client = Together()

## Basic transcription

response = client.audio.transcriptions.create(
    file=Path("audio.mp3"),
    model="openai/whisper-large-v3",
    language="en",
)
print(response.text)

## Basic translation

response = client.audio.translations.create(
    file=Path("foreign_audio.mp3"),
    model="openai/whisper-large-v3",
)
print(response.text)

​
Available models

The following speech-to-text models are available:
OrganizationModelModel string for APIServerlessDedicated
OpenAIWhisper Large v3openai/whisper-large-v3✅✅
NVIDIAParakeet TDT 0.6B v3nvidia/parakeet-tdt-0.6b-v3✅✅
NVIDIANemotron 3 ASR Streaming 0.6Bnvidia/nemotron-3-asr-streaming-0.6b✅✅
NVIDIANemotron 3.5 ASR Streaming 0.6Bnvidia/nemotron-3.5-asr-streaming-0.6b✅✅
DeepgramNova-3 (English)deepgram/nova-3-en❌✅
DeepgramNova-3 Multilingualdeepgram/nova-3-multi❌✅
DeepgramFluxdeepgram/flux❌✅
See the serverless catalog and the dedicated endpoint model catalog for pricing and additional deployment options.

​
Limits

LimitValueNotes
Max request size (direct upload)500 MBRequests above this are rejected at the edge with HTTP 413 Payload Too Large. For anything larger, host the audio at a public HTTPS URL and pass that URL as the file field instead.
Max file size (URL fetch)1 GBWhen you submit an HTTPS URL instead of binary, the server downloads up to 1 GB. Larger downloads fail with 400 file_too_large.
Max audio duration4 hours per requestLonger audio is rejected with 400 audio_too_long. Split into ≤ 4 h segments and submit separately.
Supported formats.wav, .mp3, .m4a, .webm, .flac, .ogg, .opus, .aac
For payloads above 500 MB, host the file at a public HTTPS URL and pass that URL as the file field instead of a binary upload. The 500 MB edge cap only applies to direct uploads. See Errors and troubleshooting for the full list of error codes.

​
Audio transcription

Audio transcription is speech-to-text in the same language as the source audio. 
from pathlib import Path

from together import Together

client = Together()

response = client.audio.transcriptions.create(
    file=Path("meeting_recording.mp3"),
    model="openai/whisper-large-v3",
    language="en",
    response_format="json",
)

print(f"Transcription: {response.text}")
The API supports the following audio formats:
  • .wav (audio/wav)
  • .mp3 (audio/mpeg)
  • .m4a (audio/mp4)
  • .webm (audio/webm)
  • .flac (audio/flac)
  • .ogg (audio/ogg)
  • .opus (audio/opus)
  • .aac (audio/aac)

​
Audio limits

The same limits apply to both /v1/audio/transcriptions and /v1/audio/translations:
  • Maximum duration: 4 hours. Longer audio is rejected with an audio_too_long error.
  • Binary uploads: Capped at 500 MB. Larger uploads return HTTP 413. Submit the audio via an HTTPS URL on the file field instead.
  • URL-fetched audio: Capped at 1 GB and 4 hours when you pass a public HTTPS URL as file.
For longer recordings, chunk the audio into ≤ 4 h segments and submit each chunk as a separate URL request. When sending a binary upload, put the model form field before the file field in the multipart body so the server can dispatch the request without buffering the full audio payload.

​
Input methods

​
Path object

Python
from pathlib import Path

response = client.audio.transcriptions.create(
    file=Path("recordings/interview.wav"),
    model="openai/whisper-large-v3",
)

​
File-like object

Python
with open("audio.mp3", "rb") as audio_file:
    response = client.audio.transcriptions.create(
        file=audio_file,
        model="openai/whisper-large-v3",
    )

​
Remote URL

The Python SDK doesn’t accept a string URL on file=. To transcribe a remote file, download it first.

​
Language support

Specify the audio language using ISO 639-1 language codes: 
from pathlib import Path

response = client.audio.transcriptions.create(
    file=Path("spanish_audio.mp3"),
    model="openai/whisper-large-v3",
    language="es",  # Spanish
)
Common language codes:
  • "en": English.
  • "es": Spanish.
  • "fr": French.
  • "de": German.
  • "ja": Japanese.
  • "zh": Chinese.
  • "auto": Auto-detect (default).

​
Custom prompts

Use prompts to improve transcription accuracy for specific contexts.
Prompts are supported only on Whisper-family models (for example, openai/whisper-large-v3). Other STT models (for example, nvidia/parakeet-tdt-0.6b-v3) accept the field for API compatibility but ignore it.
from pathlib import Path

response = client.audio.transcriptions.create(
    file=Path("medical_consultation.mp3"),
    model="openai/whisper-large-v3",
    language="en",
    prompt="This is a medical consultation discussing patient symptoms, diagnosis, and treatment options.",
)

​
Next steps