Quotas and limits in SpeechKit
Yandex SpeechKit has the following limits:
- Quotas are organizational restrictions that can be changed by technical support on request.
- Limits are technical limitations due to Yandex Cloud architectural features. The limits cannot be changed.
If you need more resources, contact support and tell us which quotas you need to increase and by how much.
Quotas
| Type of limit | Value |
|---|---|
| Streaming speech recognition | |
| Requests per second | 40 |
| Synchronous recognition | |
| Requests per second | 20 |
| Asynchronous recognition | |
| Recognition requests per hour | 500 |
| Operation status check requests via API v2 per hour | 2,500 |
| Operation status check requests via API v3 per second | 5 |
| Billable hours of audio per day1 | 10,000 |
| Queries accessing an LLM | |
| Concurrent queries accessing generative text models | 2 |
| Speech synthesis | |
| Requests per second | 40 |
1 The first recognition request triggers the start of the time count.
Limits
| Type of limit | Value |
|---|---|
| Streaming speech recognition | |
| Maximum duration of transmitted audio for entire session | 5 minutes |
| Maximum size of transmitted audio data | 10 MB |
| Maximum number of audio channels | 1 |
| Synchronous recognition | |
| Maximum file size | 1 MB |
| Maximum duration of audio | 30 seconds |
| Maximum number of audio channels | 1 |
| Asynchronous recognition | |
| Maximum file size when uploading to a bucket | 1 GB |
| Maximum file size when uploading in an API v3 request body | 60 MB |
| Maximum duration of audio | 4 hours |
| Period for storing recognition results on the server | 3 days |
| Speech synthesis | |
| Minimum duration of a pattern for synthesis | 1 second |
| Maximum request size for the API v1 | 5,000 characters |
| Maximum request size for the API v3 | 250 characters and 24 seconds |
| Maximum request size for the API v3 in unsafe mode | 5,000 characters |
| Maximum request size for the API v3 in streaming mode | 5,000 characters |
| Queries accessing an LLM | |
| Number of instructions for the generative text model per session | 16 |