The Audio API supports two speech to text endpoints:

• transcriptions

• translations

To get started with the Audio API, please read our speech to text developer documentation.

FAQ

How much does the Audio API cost to use?

See our pricing page for details.

What languages are supported?

View a list of supported languages here.

How can we handle large audio files?

For legacy/whisper-1 Audio API transcription uploads, the maximum request size is 25 MiB. Newer gpt-4o transcription routes may use different validation, such as duration or token limits, so check the model-specific documentation when handling long audio inputs from users.

What streaming methods are available?

There are two ways you can stream your transcription depending on whether you are transcribing a completed audio recording or an ongoing stream of audio:

• Streaming the transcription of a completed audio recording

• Streaming the transcription of an ongoing audio recording

For ongoing transcription with gpt-live-transcribe, your application must detect when speech ends and send input_audio_buffer.commit to signal the end of each turn and receive the final transcript. This model does not support server_vad or semantic_vad.

Note that streaming is not supported with the whisper-1 model.

What file formats are supported?

The supported file formats are included in our API docs.

Can I send links to audio files to the Audio API?

No, you must send a file in one of the supported audio formats.