Skip to main content
Cohere

Search documentation

Type to search this documentation.

On this pageOverview

Audio Transcription - quickstart

The Audio Transcriptions API provides a dedicated speech-to-text endpoint for uploaded audio files. It features:

  • Performant audio transcription with low word error rate (WER)
  • A simple multipart upload flow
  • Multilingual support
  1. Prerequisites

  2. Setting up the audio file

    The Audio Transcriptions endpoint accepts FLAC, MP3, MPEG, MPGA, OGG, and WAV files of 25MB or less. For this quickstart, we'll use the following WAV example.

    If you don't have an audio file already, download the example to follow along.

  3. Passing the audio file

    Use the following code to pass the audio file to the Audio Transcription endpoint.

    curl
    curl -X POST "https://api.cohere.com/v2/audio/transcriptions" \
      -H "Authorization: Bearer $TRIAL_KEY" \
      -F "model=cohere-transcribe-03-2026" \
      -F "language=en" \
      -F "file=@./transcribe-model-sample-derrida-mashup.wav" \

    For Cohere Transcribe Arabic, use the following snippet. Note: the sample audio in this quickstart is in English, so language is set to en. For audio files in which the predominant language is Arabic, make sure to set language to ar.

    curl
    curl -X POST "https://api.cohere.com/v2/audio/transcriptions" \
      -H "Authorization: Bearer $TRIAL_KEY" \
      -F "model=cohere-transcribe-arabic-07-2026" \
      -F "language=en" \
      -F "file=@./transcribe-model-sample-derrida-mashup.wav" \

    Install the latest Cohere SDK.

    Bash
    pip -U install cohere

    Import the ClientV2 class, instantiate it, and then use its create method to pass the audio file.

    Python
    from cohere import ClientV2
    import os
    
    api_key = os.environ["TRIAL_KEY"]
    client = ClientV2(api_key)
    
    with open("transcribe-model-sample-derrida-mashup.wav", "rb") as f:
        response = client.audio.transcriptions.create(
            model="cohere-transcribe-03-2026", language="en", file=f
        )
    
    print(response)

    For those using Cohere Transcribe Arabic, this step becomes the following. Note: the sample audio in this quickstart is in English, so language is set to en. For audio files in which the predominant language is Arabic, make sure to set language to ar.

    Python
    from cohere import ClientV2
    import os
    
    api_key = os.environ["TRIAL_KEY"]
    client = ClientV2(api_key)
    
    with open("transcribe-model-sample-derrida-mashup.wav", "rb") as f:
        response = client.audio.transcriptions.create(
            model="cohere-transcribe-arabic-07-2026",
            language="en",
            file=f,
        )
    
    print(response)
  4. Receiving the response

    It should return a response similar to the following.

    Bash
    {"text":"I speak only one language, and it's not my own, but the poet is a man of metaphor."} 
    Bash
    text="I speak only one language, and it's not my own, but the poet is a man of metaphor."
Suggest an edit

Propose a replacement for this page. The site team reviews it before applying any changes.

Export
Documentation menu