Base Model 20250808 Internal Error when training with Structure Text file

Hello Kaiyip Ho,

Welcome to the Microsoft Q&A and thank you for posting your questions here.

I understand that you are experiencing an Internal Error when training an Azure AI Speech Custom Speech model with Structured Text and output-formatting files using the English base model20250808.

With information provided the issue is not caused by the Structured Text file itself, because the same validated files train successfully with base model 20221013, but fail repeatedly with base model 20250808. Microsoft documentation still lists Structured Text as a supported Custom Speech training dataset type, this points to a base-model-specific backend regression, undocumented compatibility limitation, or service-side training issue for 20250808, not a normal customer file-format problem. - https://learn.microsoft.com/en-us/azure/ai-services/speech-service/how-to-custom-speech-test-and-train, and https://learn.microsoft.com/en-us/azure/ai-services/speech-service/how-to-custom-speech-train-model

The below steps will resolve the issue:

• Stop modifying already validated Structured Text files without a specific validation error.
• Use base model 20221013 temporarily if it meets the required recognition quality and lifecycle needs.
• Create a minimal reproduction dataset that succeeds on 20221013 but fails on 20250808.
• Collect the failed training job details, dataset IDs, region, locale, UTC timestamp, screenshots, and correlation or request information.
• Open an Azure technical support request under Azure AI services > Speech > Custom Speech > Training/Fine-tuning or use Priority Customer Support (PCS): https://learn.microsoft.com/en-us/azure/azure-portal/supportability/priority-community-support and request product-team confirmation on whether Structured Text adaptation is supported for base model 20250808. - https://learn.microsoft.com/en-us/azure/azure-portal/supportability/how-to-create-azure-support-request, OR use - https://azure.microsoft.com/en-us/support/

This is the only reliable path to both continue the project safely and obtain the authoritative root cause from Microsoft engineering. Use the following official Microsoft resources for more reading and implementation steps:

• https://learn.microsoft.com/azure/ai-services/speech-service/how-to-custom-speech-test-and-train
• https://learn.microsoft.com/azure/ai-services/speech-service/how-to-custom-speech-train-model
• https://learn.microsoft.com/azure/ai-services/speech-service/how-to-custom-speech-create-project
• https://learn.microsoft.com/azure/ai-services/speech-service/customize-pronunciation
• https://learn.microsoft.com/azure/ai-services/speech-service/rest-speech-to-text
• https://learn.microsoft.com/azure/azure-portal/supportability/how-to-create-azure-support-request

I hope this is helpful! Do not hesitate to let me know if you have any other questions, steps or clarifications.

Please don't forget to close up the thread here by upvoting and accept it as an answer if it is helpful.

Hi @Kaiyip Ho

This looks like a model/version-specific issue rather than a problem with your data.

Since the same structured text works with an older base model (20221013) but fails with 20250808, it likely indicates a limitation or backend issue with the newer model build. Similar “internal error / 500” behaviors are typically related to service-side issues or unsupported scenarios rather than file validation problems. [community.openai.com]

You can try the following:

Use the older base model temporarily (as you already confirmed it works)

Retry the training job after some time to rule out transient backend issues

Keep checking if there is any update on supported formats for the newer model

At the moment, it doesn’t look like an issue with your structured text file itself.

Hope this helps!

Thankyou!