StyleTTS: Reference Audio ineffective

[andrewhowdencom]

Holaa! I hope you're doing well.

Am trying to get the StyleTTS model working, but cannot for the life of me get it to change at all based on the "reference audio". Without getting into too many details, snippet is here:

        # Check to see if the file in the voices path exists before starting the model. Otherwise, the model falls back
        # to a "default voice" somehow.
        p = os.path.join(self.reference_audio_path, f"{voice}.wav", )
        if not os.path.isfile(p):
            raise Exception(f"Voice file not found: {p}")

        voice = StyleTTSVoice(
            model_config_path=self.model_config_path,
            model_checkpoint_path=self.model_checkpoint_path,
            ref_audio_path=p
        )

        engine = StyleTTSEngine(
            style_root=self.styletts_checkout,
            diffusion_steps=15,
            voice=voice,
        )

        stream = TextToAudioStream(
            engine=engine,
            tokenizer="stanza",
            muted=True,
        )

        with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as tmp_wav:  # Create temp file for result
            stream.feed(_split_paragraph(sys.stdin.read())).play(
                output_wavfile=tmp_wav.name,
            )

        engine.shutdown()

        # Process the input wav into something that the output can use. The input filename is passed directly to ffmpeg,
        # implicitly allowing the user to set formats (e.g. mp3)
        (
            ffmpeg.input(tmp_wav.name)
                .output(to)
                .overwrite_output()
                .run()
        )

Is there something obvious I am missing? It all seems to work similarly to another StyleTTS version (NeuralVox), but I can't quite connect the dots here.

Is there a working example of this?