Different kinds of inference

[rose-jinyang]

Hello
How are you?
I found two kinds of inference for the offline model in your demo page.

I checked the script "run_voice_clone.py" but I am NOT sure how I can do two kinds of inference in the script.
In your demo page, ONLY a prompt speech is inputted for Reference Utterance as Prompt.
In the script "run_voice_clone.py", both prompt text and prompt audio are inputted.
Besides, the limitation for lengths of prompt text or audio are NOT mentioned.

[Paulmzr]

@rose-jinyang

Yes, if you need to use the voice clone script I provided, you must know the corresponding audio for your reference audio. If you wish to use a 3-second prefix, you can treat that 3-second prefix as the reference audio and use the input to be synthesized as the complete input (left side of Line 109).

SLED-TTS/scripts/run_voice_clone.py

Line 109 in b8ed10d

input_text = [prompt_text + " " + input_text]

[rose-jinyang]

Hi @Paulmzr
First, let's see the case using Reference Utterance as Prompt.
In your demo page, ONLY prompt audio as prompt data is used .

But in your script "run_voice_clone.py", both a reference audio and the corresponding prompt text are used.