Azure AI Speech
An Azure service that integrates speech processing into apps and services.
1,675 questions
Training custom model in Speech Studio results in "Internal error."
![](data:image/svg+xml, %3Csvg xmlns='http://www.w3.org/2000/svg' height='64' class='font-weight-bold' style='font: 600 30.11764705882353px "SegoeUI", Arial' width='64'%3E%3Ccircle fill='hsl(128, 61%, 22%)' cx='32' cy='32' r='32' /%3E%3Ctext x='50%25' y='55%25' dominant-baseline='middle' text-anchor='middle' fill='%23FFF' %3EAO%3C/text%3E%3C/svg%3E)
Asger Ougaard
0
Reputation points
I'm trying to train a custom Speech-to-Text model in Azure AI Speech Studio for Danish speech.
I have made a structured text phonetic pronunciation markdown file.
The markdown file uploads without errors as a speech dataset. However, when I try to train a custom model using this file the status says "Processing" for a few days and then fails with the enigmatic "Internal error." message.
I have tried to simplify the markdown file, reducing it to only a few simple words to see if that would help, but the custom model training still fails with the same error message.
I have created a similar file for English speech and successfully trained a custom model.
I am at a loss for what to do now as the error message doesn't help with the troubleshooting and every step up until the training seems to be correct. Has anyone experienced similar problems? Am I missing something?
{count} votes
-
romungi-MSFT 44,771 Reputation points Microsoft Employee2024-08-27T12:26:58.8266667+00:00 @Asger Ougaard I have recently worked with another user who faced similar issue. Their training failed with internal error without many inputs on troubleshooting while it worked for sample dataset they used from documentation. Eventually, it turned out to be an issue with the audio file which turned out to be of longer duration. The documentation recommends using small datasets with a limit on overall hours of audio. In your case where the training worked for english, could you also use similar data length or audio time for training with Danish speech? Thanks!!
-
Asger Ougaard 0 Reputation points
2024-08-28T07:34:22.48+00:00 @romungi-MSFT I am using structured text dataset, no audio data. Documentation says up to 10 classes, 4,000 items and 50,000 training sentences. My dataset currently has 2 classes and consists of 129 lines of text.
It says training with text should take minutes, but the processing hangs for days without failing. -
romungi-MSFT 44,771 Reputation points Microsoft Employee
2024-09-02T13:21:04.93+00:00 @Asger Ougaard Apologies for the delay. Unfortunately, for this type of error message it would need the support team's help to check for any issues with data while processing the files during training. Do you have a support subscription to raise a case?
Sign in to comment
Sign in to answer