Hello Team,
Kindly need your help on the issue facing while using real time transcription with custom domain in azure Speech service with ASR Engines. Please find the below error message.
ERROR: Cancelled: ErrorCode=ServiceTimeout,ErrorDetails=Due to service inactivity, the client buffer exceeded maximum size. Resetting the buffer
Also, I have tried few suggestions and requesting for your insights for below scenarios.
- Review Configuration Settings:
- Kindly review thoroughly the configuration settings for buffer size and buffer management policies. It’s possible that there’s a misconfiguration that is contributing to the problem. Adjusting the buffer size or modifying the rules for data management during inactive periods may be necessary.
[We are not configuring anything unusual with the buffer size and buffer management policies. It is default as is. Generally, it’s been working fine until lately.
Do you have any recommended buffer settings, for our consideration? Can this configuration be applied in the property settings of the Azure Speech SDK’s ASR connection?]
- Prevent Unnecessary Session Openings:
- Ensure that sessions are not left open unnecessarily or without activity. If a session remains idle for an extended period, it could lead to buffer overflow and subsequent timeouts. Kindly close sessions promptly when they are no longer needed.
[Well, as per requirement, we are intentionally opening ASR connections with real-time streaming for an average of 8 hours each.
We will stop the connections once it is no longer required.
There will be times where there will be streaming of silent audio (no speech) for a certain period of time, perhaps an hour.
Would there be an expected ASR error of ServiceTimeout when we are streaming silent or no-speech audio?]
- Implement Keep-Alive Messages:
- To prevent timeouts during expected inactivity, consider implementing keep-alive messages. These periodic messages can maintain an active connection even when no other data is being transmitted. This will help prevent the buffer from reaching its maximum size.
[we are still streaming silent/no-speech audio which we consider it as data. Unless Azure Speech Service considers only speech audio as data?
Kindly elaborate further on the proposed keep-alive messages method. Is it another type of data, other than audio signal that we can send in an open ASR connection?
It would be very helpful if you could please share any Microsoft document for the ServiceTimeout error and for above suggestion.
Thank you in advance.