Regression: TTS Journey voices with Opus: "This voice currently only supports LINEAR16 and MULAW output".

[chrbsg]

Client

Go v1 SDK but looks like a backend server regression/bug.

Environment

Linux

Code and Dependencies

        req := texttospeechpb.SynthesizeSpeechRequest{
                Voice: &texttospeechpb.VoiceSelectionParams{
                        LanguageCode: "en-US",
                        Name:         "en-US-Journey-D",
                },
                AudioConfig: &texttospeechpb.AudioConfig{
                        AudioEncoding:   texttospeechpb.AudioEncoding_OGG_OPUS,
                        SampleRateHertz: 48000,
                },
        }
        req.Input = &texttospeechpb.SynthesisInput{
                        InputSource: &texttospeechpb.SynthesisInput_Ssml{
                                Ssml: "hi",
                        },
                }
        resp, err := client.SynthesizeSpeech(ctx, &req)

Expected behavior

Opus audio should be generated.

Actual behavior

This code has been working for a long time (since Journey voices were first introduced), but today the client.SynthesizeSpeech call is returning an error:

code = InvalidArgument 
desc = This voice currently only supports LINEAR16 and MULAW output.

Additional context

This looks a lot like the transcription regression where Opus support was apparently removed accidentally: googleapis/google-cloud-node#5609 @danielbankhead

[codyoss]

This sounds like a backend regression and not related to the client library itself. I would recommend reaching out from the service support page to report the issue: https://cloud.google.com/speech-to-text/docs/support

[chrbsg]

I opened issue https://issuetracker.google.com/issues/367647327 on the Cloud Platform issue tracker.

[SuppliedOrange]

As of the latest update, here's what they say:
https://cloud.google.com/text-to-speech/docs/voice-types#journey_voices_preview

Note: Journey Voices doesn't support SSML input, speaking rate and pitch audio encodings, and only returns LINEAR16 or MULAW audio.

Here's the full list of changes for future reference: https://cloud.google.com/text-to-speech/docs/release-notes

[SuppliedOrange]

Any advice with converting the resulting .wav to .mp3 is appreciated. I've tried using ffmpeg and pydub (which also uses ffmpeg underneath) but the audio turns weird and monodirectional.