Supported Formats and Limits
Learn about the supported input and output audio formats for the Flow APIInput audio
All input audio (i.e. the user's voice) sent to the Flow API must be raw PCM audio in one of the following formats:
- PCM F32 LE raw audio stream (32-bit float)
- PCM S16 LE raw audio stream (16-bit signed int)
Other audio encodings are not supported. Sample rates are not restricted, but we recommend using 16kHz.
Output audio
The Flow API will always return PCM audio in PCM S16 LE format, regardless of the input format.
The output audio sample rate is always 16kHz.
Usage Limits
The Flow API limits the number of hours of audio users can process each month to help manage load on our servers. The current limits (in hours) by account type are listed in the table below:
| Tier | Max. hours per month | Concurrent sessions |
|---|---|---|
| Free Tier | 50 | 1 |
| Paid Tier | 1000 | 1 |
| Enterprise | Custom | Custom |
Please reach out to Support if you need to increase the above limits.
Guidance for users
Clients can disconnect a session before it is automatically terminated and immediately reconnect a new session. Note that new sessions will typically start in less than a second. If seamless transition is required, the new session can be connected a few seconds before disconnecting the old session.
Since unpredictable network issues can cause WebSocket connections to be dropped, we recommend graceful handling of session termination for long-running sessions.
Data Retention
Conversation audio and transcriptions by the Flow API are not stored.