Speech to text Custom Model is failing when used lasted base model for BYOS speech resource

Ulhas Hulyal, Nilesh 0 Reputation points
2025-01-24T07:18:54.03+00:00

We are trying to train custom model from speech portal using speech service deployed with BYOS.We noticed it is failing when used latest base model , but when used old base model custom model is succeeding .

When used below old base model for English it worked

CustomModel_Failed_BaseModel

But, when used below latest base model mentioned for English it is failing

CustomModel_Failed_EarlierBaseModel

We have observed the same behavior with the Spanish language as well.

Azure AI Speech
Azure AI Speech
An Azure service that integrates speech processing into apps and services.
1,880 questions
{count} votes

1 answer

Sort by: Most helpful
  1. santoshkc 11,790 Reputation points Microsoft Vendor
    2025-01-24T15:15:18.08+00:00

    Hi @Ulhas Hulyal, Nilesh,

    Thank you for sharing the details regarding the issue you're facing with training a custom model using Speech Service deployed with BYOS. This issue is also observed in both languages, suggesting that it may be related to the deployment configuration of Speech Service with BYOS.

    Regarding the potential connection between the container support and Speech Service deployment with BYOS, Azure Blob Storage does support optional anonymous read access to containers and blobs. However, please ensure that the Speech Service has the necessary permissions to interact with the storage account and write data.

    If your Speech Service deployed with BYOS does not have the required permissions, this could be causing the issue. Please verify that the Speech Service's managed identity is granted the appropriate roles where the models are stored and ensure there are no restrictions blocking access.

    Please refer to this for more info: Set up the BYOS Speech resource.

    Thank you for your understanding.

    0 comments No comments

Your answer

Answers can be marked as Accepted Answers by the question author, which helps users to know the answer solved the author's problem.