Limits
This page lists the limits of each audio request and session, and what you get when you go over one. Request rates and monthly quotas are workspace settings, covered in Rate limits and quotas.
Transcription
Transcription
These limits apply to file transcription and realtime sessions:
| Limit | Value | Over the limit | Details |
|---|---|---|---|
Audio URL length (file_url) | 2,083 characters | Validation error | File transcription |
Uploaded file (file_id) | Upload rejected | File transcription | |
Context window, shared by the audio and context_bias | 65,536 tokens | 400 error, nothing is cut | Limits and errors |
| Timestamps | Per segment only: timestamp_granularities=segment | word returns a 400 error | Timestamps |
| Speaker references | Up to 8 clips, 3 to 15 seconds each, with diarize set to true | The request is rejected with a 400 error: "At most 8 speaker references are supported, got {n}". | Name speakers with reference clips |
Streaming delay (target_streaming_delay_ms) | Any integer from 80 to 1200, or 2400 | error event | Control latency |
Speech generation
Speech generation
These limits apply to the values a speech request accepts:
| Limit | Value | Over the limit | Details |
|---|---|---|---|
expressiveness | 422 validation error | Expressiveness | |
Phonemes in phoneme_hints | A closed list of 50 phonemes | 422 error listing the accepted phonemes | Phonemes |
| Emotion tags | 34 values | No error: the tag isn't spoken, and has no documented effect | Supported emotions |
| Pace tags | slow or fast | No error: the tag isn't spoken, and has no documented effect | Pace tags |
| Voice cloning reference clip | A few seconds to a few minutes | 400 error for a file that isn't MP3 or WAV | Prepare a reference clip |
Audio in chat completions
Audio in chat completions
This limit applies to chat requests that carry audio inline:
| Limit | Value | Over the limit | Details |
|---|---|---|---|
Request body with input_audio | 429 error, Rate limit exceeded | Audio in chat completions |
Voice agents
Voice agents
Webhook deliveries to your endpoint follow these limits:
| Limit | Value | Details |
|---|---|---|
| Webhook response time | Your endpoint must answer within 10 seconds | Delivery and retries |
| Webhook attempts | Up to 3 in total, on 5xx and 429 responses and on network errors | Delivery and retries |
Rate limits and quotas
Rate limits and quotas
Request rates and monthly quotas are set per workspace, and shared by all its API keys. See Workspace usage and limits.
Going further