Inquiry Regarding Azure AI Speech Error

y.ashibe 25

Dear Azure Support Team

I recently encountered an issue while using Azure AI Speech service with recordings from the VoiceMemo app on iPhone. Specifically, when attempting to process recordings of approximately 30 minutes in length, I received the following error:

"errorMessage": "Invalid audio channel configuration. (JobId ***************).",
"errorKind": "InvalidChannelSpecification"

It's worth noting that shorter recordings, around 1 to 2 minutes in duration, are processed successfully without any errors.

Given this discrepancy, I suspect there might be a bug within the Azure AI Speech service related to the handling of longer recordings from the VoiceMemo app.

Could you kindly investigate this matter further and provide any insights or resolutions that may be available? Your assistance in resolving this issue would be greatly appreciated.

Thank you for your attention to this matter.

Best regards

y.ashibe 25 Reputation points

2024-05-15T22:16:48.6666667+00:00

This comment will be deleted
y.ashibe 25 Reputation points

2024-05-15T22:17:51.7133333+00:00

This comment will be deleted
navba-MSFT 18,160 Reputation points Microsoft Employee

2024-05-16T08:22:59.8166667+00:00
@y.ashibe Welcome to Microsoft Q&A Forum, Thank you for posting your query here!

Could you please provide details about your use case and what you are trying to achieve ?

Are you using it for Speech to Text?

Is it using Batch Transcription API in the background? If yes, what is the audio format it is using?

The batch transcription API supports many different formats and codecs, such as:

WAV

MP3

OPUS/OGG

FLAC

WMA

AAC

ALAW in WAV container

MULAW in WAV container

AMR

WebM

M4A

SPEEX

4.Also enable the speech SDK logging to check if there are any errors at the time of the issue:

https://learn.microsoft.com/en-us/azure/ai-services/speech-service/how-to-use-logging#sample

5.We would need the sessionID and time stamp of the issue and the SDK logs to troubleshoot this issue further.

6.Please share the audio file also with us so that we can repro this issue at our end.

Awaiting your reply.
y.ashibe 25 Reputation points

2024-05-17T01:02:32.23+00:00
@navba-MSFT
Thank you very much for your assistance.

We have developed a business chat system that automatically transcribes audio files attached to chats from users.
In this case, the problem occurred when a file recorded with an iPhone voice memo was posted to a chat session using our application on the iPhone.
The server is configured in Java, and the attached file is uploaded to Storage and then sent to Azure AI Speech's Speech to Text batch transcription API (POST {endpoint}/speechtotext/v3.1/ transcriptions).
Recording with an iPhone is a very major case and we are currently having a very difficult time providing service to our clients due to this outage.

Yes

Batch transcription API.
The audio format is m4a.
iPhone voice memo is used. I have tried both lossless and lossy compression methods, but the result is the same: an error.
If I convert from m4a to mp3, transcription was possible, but due to the ACC license, I would like to transcribe as m4a without converting to mp3.

The processing is done in Java, but for various reasons, a REST API is used.
{
"successfulTranscriptionsCount": 0,
"failedTranscriptionsCount": 1,
"details": [
{
"source": "https://********.blob.core.windows.net/*****.m4a",
"status": "Failed",
"errorMessage": "Invalid audio channel configuration. (JobId ce8722dd-7c69-4f81-a3c2-5c5a70817d9f).",
"errorKind": "InvalidChannelSpecification"
}
]
}

We will also try the SDK, but it will take some time since it involves source modification.
If the REST API can also output detailed logs, we would appreciate it if you could let us know.
I will provide the information above for the time being.

https://drive.google.com/file/d/1wD8WCPD8OlvEU6yFi38HRsyfZH11fp7d/view?usp=sharing

This community is uploading to GoogleDrive due to the 3MB limit.

The conditions are as follows
API：https://japaneast.api.cognitive.microsoft.com/speechtotext/v3.1/transcriptions
locale：ja-JP
properties->wordLevelTimestampsEnabled:false
->diarizationEnabled:false
->languageIdentification->candidateLocales:ja-JP,en-US

The same was true with the Whisper model in eastus v3.2preview.
y.ashibe 25 Reputation points

2024-05-17T05:13:31.48+00:00

https://learn.microsoft.com/en-us/azure/ai-services/speech-service/batch-transcription-create?pivots=rest-api

SDK class for batch transcription not found.
navba-MSFT 18,160 Reputation points Microsoft Employee

2024-05-17T05:39:59.97+00:00
@y.ashibe Thanks for confirming that you are not using SDK and using the Batch REST API.

While the transcription fails, in the REST API response headers, it will provide the detailed error.

Please share the below details in the private message, this will help me to look at the backend logs and identify the cause of the issue:

Failing Job ID and Session ID.

Complete resource URI of the speech service in below format:
/subscriptions/XXXXX/resourceGroups/XXXX/providers/Microsoft.CognitiveServices/accounts/XXXXXX

Time of the issue in UTC.

Region where the speech resource is deployed.

Awaiting your reply.
y.ashibe 25 Reputation points

2024-05-20T07:31:34.0033333+00:00

@navba-MSFT

We have sent you a private message. Please confirm.
navba-MSFT 18,160 Reputation points Microsoft Employee

2024-05-20T07:34:10.74+00:00

@y.ashibe I have got your message. I have involved the Product owners. I will keep you posted once I hear from them.
y.ashibe 25 Reputation points

2024-05-23T05:13:38.1233333+00:00

@navba-MSFT

Thank you very much, this was very helpful.

I have confirmed that I can transcribe the moov atom by moving it forward using the “-movflags faststart” command in ffmpeg.
However, it is very disappointing that the data recorded with the iPhone cannot be transcribed as is.

I may not be able to answer this question, but is there any possibility that the subsystem will be improved in the future so that data over 10MB with moov atom backward can be read?
Since you have listed m4a in your support file, it would be better if your company can handle this.

I will send this to you as a request.
navba-MSFT 18,160 Reputation points Microsoft Employee

2024-05-23T05:57:57.8966667+00:00

@y.ashibe Thanks for getting back. I will pass on this feedback to the Product Owners to check the feasibility on supporting this.

In the meantime, I would also suggest you to post this feedback request in our Azure feedback forum which will be monitored by the Product owners:

Azure Speech Service feedback forum: https://feedback.azure.com/d365community/forum/09041fae-0b25-ec11-b6e6-000d3a4f0858

Also for the below answer, please do not forget to "Accept the answer” and “up-vote” wherever the information provided helps you, this can be beneficial to other community members.

Accepted answer

navba-MSFT 18,160 Reputation points Microsoft Employee

2024-05-23T02:53:10.1233333+00:00

@y.ashibe Apologies for the late response. I appreciate your patience on this.

I have got a reply from the Product owners. They have analyzed the backend logs based on the details which you had shared.

.

Cause:

The problem with the file is that the metadata is at end of the file. The concrete decoding error we get is that no moov atom can be found within the first 10MB. If the audio encoding subsystem cannot determine the format we cannot process it. This should explain why 1-2m files work since the file is below 10MB, hence the metadata being found.

.

Action Plan:

The recommended way to fix the issue is the preprocess the audio, e.g., with ffmpeg.

.

If you have any follow-up questions, please feel free to ask. I would be happy to answer it.
Please sign in to rate this answer.

1 person found this answer helpful.
navba-MSFT 18,160 Reputation points Microsoft Employee

2024-05-23T05:58:51.43+00:00

@y.ashibe Thanks for getting back.

.

I will pass on this feedback to the Product Owners to check the feasibility on supporting this.

.

In the meantime, I would also suggest you to post this feedback request in our Azure feedback forum which will be monitored by the Product owners:

Azure Speech Service feedback forum: https://feedback.azure.com/d365community/forum/09041fae-0b25-ec11-b6e6-000d3a4f0858

.

Also for the above answer, please do not forget to "Accept the answer” and “up-vote” wherever the information provided helps you, this can be beneficial to other community members.

y.ashibe 25 Reputation points

2024-05-24T00:42:18.5766667+00:00

@navba-MSFT

Thank you!

I have sent you a FEDBACK.

https://feedback.azure.com/d365community/idea/02d4245b-6619-ef11-989a-000d3ae2b698

y.ashibe 25 Reputation points

2024-05-24T00:43:57.6733333+00:00

@navba-MSFT

Thank you!

I have sent you a FEDBACK.
https://feedback.azure.com/d365community/idea/02d4245b-6619-ef11-989a-000d3ae2b698

navba-MSFT 18,160 Reputation points Microsoft Employee

2024-05-25T03:02:26.3666667+00:00

@y.ashibe If its not too much to ask, Could you please copy and paste the same feature request in the below channel too to keep a track of the progress?

https://github.com/cognitive-services/media-transform/issues

The Product owners have direct visibility here.

Thanks again.

y.ashibe 25 Reputation points

2024-05-27T00:33:06.4166667+00:00

delete

y.ashibe 25 Reputation points

2024-05-27T01:26:55.44+00:00

@navba-MSFT

Sorry, but I don't seem to be authorized to do this.

Maybe GitHub is not a public repository.

navba-MSFT 18,160 Reputation points Microsoft Employee

2024-05-27T03:21:06.4666667+00:00

@y.ashibe Yes, I just submitted the ask in above GitHub channel.
https://github.com/cognitive-services/media-transform/issues/247

Awaiting reply from Product owners.
Sign in to comment

Share via

Inquiry Regarding Azure AI Speech Error

0 additional answers