I am trying to finetune and get better results from Azure Speech to Text

Question

I am trying to finetune and get better results from Azure Speech to Text

Farida Bharmal 0

I am using Azure AI Playground to train a model for speech recognition. I have the audio files that I need to get accurate transcription. However, when I upload my data (audio file zip), I get error "the dataset is not suitable for this operation"

Pavankumar Purilla 7,100 Reputation points Microsoft External Staff Moderator

2025-05-05T19:00:02.6366667+00:00

Hi Farida Bharmal,

When encountering the error message “The dataset is not suitable for this operation” while training an Azure AI Speech recognition model, it typically indicates that the audio files do not meet the platform’s strict requirements. Ensuring compliance with these specifications is crucial for successful model training. Each audio file must follow the RIFF (WAV) format, have a sample rate of either 8,000 Hz or 16,000 Hz, be mono-channel, and maintain a maximum length of 40 seconds per file. Additionally, files must be encoded in PCM, 16-bit format and packed within a .zip archive not exceeding 2 GB or 10,000 files. To validate these files, SoX is a useful command-line tool that can confirm file properties and assist in conversions.

For example, you can check file format and properties with the command sox --i your_audio_file.wav, or convert a file using sox input.mp3 -b 16 -e signed-integer -c 1 -r 16k -t wav output.wav. Common issues include files exceeding the length limit, oversized ZIP archives, and incorrect format settings—each of which can be resolved by segmenting audio files, creating multiple ZIP archives, or converting stereo files to mono with the correct sample rate. Best practices include testing with a small dataset, ensuring high-quality recordings, and incorporating diverse speakers to enhance model effectiveness.

For more information: Training and testing datasets - Speech service - Azure AI services | Microsoft Learn

I hope this information helps. Thank you!
Pavankumar Purilla 7,100 Reputation points Microsoft External Staff Moderator

2025-05-06T21:38:46.5866667+00:00

Hi Farida Bharmal,
Did you get any chance to check the response. Thank you!

1 answer

Your answer

Pavankumar Purilla 7,100 Reputation points Microsoft External Staff Moderator

2025-05-05T19:00:02.6366667+00:00

Hi Farida Bharmal,

When encountering the error message “The dataset is not suitable for this operation” while training an Azure AI Speech recognition model, it typically indicates that the audio files do not meet the platform’s strict requirements. Ensuring compliance with these specifications is crucial for successful model training. Each audio file must follow the RIFF (WAV) format, have a sample rate of either 8,000 Hz or 16,000 Hz, be mono-channel, and maintain a maximum length of 40 seconds per file. Additionally, files must be encoded in PCM, 16-bit format and packed within a .zip archive not exceeding 2 GB or 10,000 files. To validate these files, SoX is a useful command-line tool that can confirm file properties and assist in conversions.

For example, you can check file format and properties with the command sox --i your_audio_file.wav, or convert a file using sox input.mp3 -b 16 -e signed-integer -c 1 -r 16k -t wav output.wav. Common issues include files exceeding the length limit, oversized ZIP archives, and incorrect format settings—each of which can be resolved by segmenting audio files, creating multiple ZIP archives, or converting stereo files to mono with the correct sample rate. Best practices include testing with a small dataset, ensuring high-quality recordings, and incorporating diverse speakers to enhance model effectiveness.

For more information: Training and testing datasets - Speech service - Azure AI services | Microsoft Learn

I hope this information helps. Thank you!
Pavankumar Purilla 7,100 Reputation points Microsoft External Staff Moderator

2025-05-06T21:38:46.5866667+00:00

Hi Farida Bharmal,
Did you get any chance to check the response. Thank you!

Answer 1

JAYA SHANKAR G S 2,525 Microsoft External Staff Moderator

Hello @Farida Bharmal ,

Usually, the error the dataset is not suitable for this operation occurs if Maximum length per audio is not satisfied.

So, if you are trying

Audio data for training or testing make sure the length of the audio per file not exceeding 60 seconds for training and 2 hours for testing
Audio + human-labeled transcript data for training or testing make sure the length of the audio per file is Two hours (testing) / 40 s (training).

Please check this and let us know if you have any query.

Thank you

JAYA SHANKAR G S 2,525 Reputation points Microsoft External Staff Moderator

2025-05-09T09:20:39.73+00:00

Hello @Farida Bharmal ,

Did you check the above answer? please check and let us know if you have any query.

Thank you
JAYA SHANKAR G S 2,525 Reputation points Microsoft External Staff Moderator

2025-05-12T03:39:01.3433333+00:00

Hello @Farida Bharmal ,

I believe above answer helped you, let us know if you have any query.

Thank you

Share via

I am trying to finetune and get better results from Azure Speech to Text

1 answer

Your answer