Windows: Calling SpeechSynthesizer.StopSpeakingAsync() does not stop synthesis

[bpasero]

Describe the bug

A call to SpeechSynthesizer.StopSpeakingAsync() does not stop synthesis for a very long time, up to 30 seconds. The log file is here: speech.log

This issue was previously reported without action at #1836 and #2264

To Reproduce

We are building a node.js binding for Speech SDK and the C++ sources mimic the samples. The synthesis is implemented here: https://github.com/microsoft/node-speech/blob/967976ce0f4887a2b5b27f486e5209a51588516f/src/main.cc#L477

The call to StopSpeakingAsync here: https://github.com/microsoft/node-speech/blob/967976ce0f4887a2b5b27f486e5209a51588516f/src/main.cc#L539

To reproduce from that module:

• using node.js 18.x on the system
• git clone https://github.com/microsoft/node-speech.git
• open index.ts and append the snippet [1] at the end
• from a terminal cd into the workspace and run npm i
• run node index.js

[1]

const t = createSynthesizer({
  modelPath: '<path to TTS model>',
  modelName: 'Microsoft Server Speech Text to Speech Voice (en-US, AriaNeural)',
  modelKey: '<model key>',
}, (error, result) => {
  if (error) {
    console.error(error);
  } else {
    console.log(result);
  }
});
t.synthesize(`
Now more than ever, developers are expected to build voice-enabled applications that can reach a global audience. With the same voice persona across languages, organizations can keep their brand image more consistent. To support the growing need for a single voice to speak multiple languages, particularly in scenarios such as localization and translation, a multi-lingual neural TTS voice is brought out in public preview.



This new Jenny Multilingual voice (preview), with US English as the primary/default language, can speak 13 secondary languages, each at the fluent level: German (Germany), English (Australia), English (Canada), English (Canada), Spanish (Spain), Spanish (Mexico), French (Canada), French (France), Italian (Italy), Japanese (Japan), Korean (Korea), Portuguese (Brazil), Chinese (Mandarin, Simplified).
`);
setTimeout(() => t.stop(), 5000);

Expected behavior

Calling SpeechSynthesizer.StopSpeakingAsync immediately stops synthesis.

Version of the Cognitive Services Speech SDK

1.37.0

Platform, Operating System, and Programming Language

• OS: Windows 11 (24H2)
• Hardware: ARM
• Programming language: C++

Additional context

This issue does not reproduce on macOS or Linux!

[ralph-msft]

Thanks for using the Speech SDK and filing this issue. We have been able to reproduce the issue you are seeing, and have added fixing this issue to our backlog. We will update here once we have an update.

As a temporary workaround, you may want to consider passing a null value as the AudioConfig to the SpeechSynthesizer constructor. You can then subscribe to the Synthesizing event which will be raised whenever the SDK receives new audio from the service. You can then pass this audio to your player of choice which should give you more control over when the audio playback stops. Please note however that calling StopSpeakingAsync may still stall for ~10-15 seconds due the underlying issue.

(B-7172399)

[bpasero]

Thanks, good to see it can be reproduced and I am looking forward to the fix 👍

[github-actions]

This item has been open without activity for 19 days. Provide a comment on status and remove "update needed" label.

[wtto00]

Hello, I am using version 1.37.0, and I have encountered a similar issue.

stopSpeaking does not immediately terminate the playback process; it only stops the speaker from playing.

For example, if I generate a 14-second audio and execute stopSpeaking at 10 seconds, then let speakResult = synthesizer?.speakSsml(ssml) will immediately return with speakResult?.reason=9(SPXResultReason_SynthesizingAudioCompleted) instead of 1(SPXResultReason_Canceled). Moreover, the callback registered with synthesizer?.addSynthesisCompletedEventHandler is triggered after waiting for 4 seconds, rather than the callback registered with synthesizer?.addSynthesisCanceledEventHandler.

let ssml =
          "<speak version='1.0' xml:lang='en-US' xmlns='http://www.w3.org/2001/10/synthesis' xmlns:mstts='http://www.w3.org/2001/mstts'><voice name='\(identifier)'>\(mstts)</voice></speak>"
let speakResult = try self.synthesizer?.speakSsml(ssml)
print(speakResult?.reason ?? "")

try synthesizer?.stopSpeaking()

Here is a demo repositorie: https://github.com/wtto00/flutter_azure_speech/tree/main/example

The swift code is in https://github.com/wtto00/flutter_azure_speech/blob/eb419b89fcc16903cabaa8f9820559d93ed80861/ios/Classes/AzureSpeechPlugin.swift#L294