Fast Transcription API for Azure AI Foundry randomly returns 429 server too busy :-(

Question

Currently, I rarely do fast audio transcription : https://learn.microsoft.com/en-us/azure/ai-services/speech-service/fast-transcription-create

For example, I did a request for the 1st time in 3 or 4 days, & immediately get 429 server too busy (am doing the calls using REST. My code is OK)

HTTP status code: 429 Response: {"code":"TooManyRequests","message":"Resource Exhausted"}

I resend the request immediately & hey presto, i get a successful transcription, for the exact same REST call, with the same audio & everything!

& I notice that usually i get the 429 error only the 1st time i try after a long time.

This is very frustrating. Audio files are a maximum of 31 seconds, either wav or ogg, & this 429 response occurs randomly

Any clues?

Thanks in advance!

Answer 1

Hello It is VMS,

Thanks for your question.

This may happen because when the service has been idle, it may not have sufficient resources immediately available to handle incoming requests. The first request after such an idle period can trigger the service to scale up its resources. During this scaling process, the service may temporarily reject requests, resulting in a 429 error.

To mitigate this issue, consider the following strategies:

1. Implement Retry Logic (As you have implemented currently)
2. Periodic Warm-Up Requests - Send periodic light weight request to keep it active

The above is cited from:

https://learn.microsoft.com/en-us/answers/questions/2260261/speech-to-text-s0-error-429-on-first-call

You can mark it 'Accept Answer' and 'Upvote' if this helped you

Regards,

Abiola

Answer 2

While this is understandable, this is not really an acceptable behaviour form an organization like Microsoft!

Of course, there's no documentation on this "idle period", so one has no clue how often to send "wake up" calls :-(