Azure Text To Speech docker container throws an exception with viseme

Jon Peterson 26

I'm using the Azure Text to Speech docker image (mcr.microsoft.com/azure-cognitive-services/speechservices/neural-text-to-speech:3.3.0-amd64-en-us-jennyneural). I'm passing it SSML through the dotnet SDK. When asking for viseme (via <mstts:viseme type="redlips_front" />), the container logs show there was an exception. I checked the images files and confirmed that there is not a file in that location. This works correctly when using the Azure hosted speech services.

ssml replace failed: System.TypeInitializationException: The type initializer for 'Microsoft.TTS.Ssml.SsmlConstant' threw an exception.
	System.IO.DirectoryNotFoundException: Could not find a part of the path '/onprem_start/phoneset/phoneset.json'.
	at Interop.ThrowExceptionForIoErrno(ErrorInfo errorInfo, String path, Boolean isDirectory, Func`2 errorRewriter)
    at Interop.CheckIo(Error error, String path, Boolean isDirectory, Func`2 errorRewriter)
    at Microsoft.Win32.SafeHandles.SafeFileHandle.Open(String path, OpenFlags flags, Int32 mode)
    at Microsoft.Win32.SafeHandles.SafeFileHandle.Open(String fullPath, FileMode mode, FileAccess access, FileShare share, FileOptions options, Int64 preallocationSize)
    at System.IO.Strategies.OSFileStreamStrategy..ctor(String path, FileMode mode, FileAccess access, FileShare share, FileOptions options, Int64 preallocationSize)
    at System.IO.StreamReader.ValidateArgsAndOpenPath(String path, Encoding encoding, Int32 bufferSize)
    at System.IO.File.InternalReadAllText(String path, Encoding encoding)
    at Microsoft.TTS.Ssml.SsmlConstant.InitPhoneset()
    at Microsoft.TTS.Ssml.SsmlConstant..cctor()
    --- End of inner exception stack trace ---
    at Microsoft.TTS.Ssml.Utility.StandardizeDefaultNamespace(String ssml)
    at Microsoft.TTS.Ssml.Ssml.Parse()
    at Microsoft.TTS.CustomTN.RuleEngine.NormalizeSSML(String locale, String ssmlstring)
    at Microsoft.TTS.CustomTN.RuleEngine.ExecuteSsml(String locale, String ssmlString, String& updated, ILogger logger, List`1 editItems)

User's image

dupammi 7,955 Reputation points Microsoft Vendor

2024-07-01T03:08:32.19+00:00
Hi @Jon Peterson

Thank you for using the Microsoft Q&A forum.

The exception you're encountering with the Azure Text to Speech Docker container when trying to use visemes (<mstts:viseme type="redlips_front" />) suggests that the container is missing a required resource or file. Specifically, it seems the container expects a phoneset.json file in a directory that doesn't exist (/onprem_start/phoneset/phoneset.json). This issue doesn't occur with the Azure hosted speech services because those services have access to the necessary resources that the Docker container might not have.

In short, it is likely due to missing files or improper paths. By ensuring the necessary files are available and correctly referenced within the container, you can resolve this error. See below:

Run the Docker container with a bash shell to explore its filesystem.

Check if the directory /onprem_start/phoneset/ exists. If not, create it and place the phoneset.json file there.

If you have phoneset.json on your host system, you can mount it into the container.

I hope you understand. Thank you.
Jon Peterson 26 Reputation points

2024-07-01T12:41:18.7566667+00:00

@dupammi you just repeated what I already mentioned in my question slightly differently. It was completely unhelpful.

There is no mention of phoneset in any of the documentation for these containers.
dupammi 7,955 Reputation points Microsoft Vendor

2024-07-02T11:05:01.8133333+00:00

Hi @Jon Peterson

Sorry for the trouble you are facing. It is possible that you are experiencing some technical glitch. To assist you with this, we recommend you provide feedback by either clicking on the smiley icon on top right of the portal as shown in below screenshot1:

Our technical team will review the information you provide and work to resolve the issue. Thank you for your cooperation.
Jon Peterson 26 Reputation points

2024-07-02T14:39:55.5466667+00:00

I'm still not sure that you fully understand the issue. The official, Microsoft Azure text to speech docker images are missing a file that seems to be required to make the viseme SSML feature work. I put in a feedback request per your suggestion, but I'm not sure if that's where this bug report should go.

I also pulled the 3.3.0-amd64-en-us-arianeural image, and it has the same problem.

Share via

Azure Text To Speech docker container throws an exception with viseme