How to use reference audio?

[Touch-Night]

I uploaded ref audio and it seems does not work.

[anden007]

me too..........

[AnyaCoder]

Sorry for not noticing your message earlier. To use reference audio, first go to the "Basic Setup" page. In the "Reference Audio" section, click the "upload" button, then upload one or more audio files (such as .mp3 or .wav formats) along with a corresponding .lab file of the same name via the file explorer. The .lab file should simply contain the transcribed text of the reference audio. The results can be seen as demonstrated in the "Basic Setup" section of the README.

[Touch-Night]

Is the .lab file essential to use reference audio? I mean it is not when use the fish-speech gradio UI.

[AnyaCoder]

In the fish-speech Gradio UI, you typically need to provide the reference text, which is essentially the content that a .lab file would contain.

[Touch-Night]

But it works if I don't.

[bunter96]

In the fish-speech Gradio UI, you typically need to provide the reference text, which is essentially the content that a .lab file would contain.

can you make permanent solution to this problem? as we know the most rich feature of fish speech is its voice cloning. I also attracted toward fish speech due to this functionality.

Possible workaround 1 : Make changes in GUI and add .lab file option so that user separately upload .lab file
Possible workaround 2 : Provide text input field beside reference audio upload and when the request made to server this written text by user will be treated as a .lab

These are my suggestions that may be not feasible. Hope u have better workaround. And thanks for this beautiful and rich functionality based GUI.

[AnyaCoder]

But it works if I don't.

def _get_pre_files(self):
        return [f for f in self.ref_files if not f.endswith(".lab")]

    def _filter_audio_files(self, pre_files: List[str]):
        return [
            f
            for f in pre_files
            if Path(f).exists() and Path(f).with_suffix(".lab").exists()
        ]

Okay, i forgot this. Seems like it will automatically handle .lab files.

[AnyaCoder]

In the fish-speech Gradio UI, you typically need to provide the reference text, which is essentially the content that a .lab file would contain.

can you make permanent solution to this problem? as we know the most rich feature of fish speech is its voice cloning. I also attracted toward fish speech due to this functionality.

Possible workaround 1 : Make changes in GUI and add .lab file option so that user separately upload .lab file Possible workaround 2 : Provide text input field beside reference audio upload and when the request made to server this written text by user will be treated as a .lab

These are my suggestions that may be not feasible. Hope u have better workaround. And thanks for this beautiful and rich functionality based GUI.

You can put .lab files besides the audio files.