app.py

import io
import os
import tempfile
from typing import List

import TTS.api
import torch
from pydub import AudioSegment
import gradio as gr  # Gradio库

import config

device = "cuda" if torch.cuda.is_available() else "cpu"

models = {}
for id, model in config.models.items():
    models[id] = TTS.api.TTS(model).to(device)


def synthesize_tts(
    text: str = 'Hello, World!',
    speaker_wavs: List[gr.File] = None,
    speaker_idx: str = 'Ana Florence',
    language: str = 'ja',
    temperature: float = 0.65,
    length_penalty: float = 1.0,
    repetition_penalty: float = 2.0,
    top_k: int = 50,
    top_p: float = 0.8,
    speed: float = 1.0,
    enable_text_splitting: bool = True,
):
    temp_files = []
    try:
        if speaker_wavs:
            # Process each uploaded file
            for speaker_wav in speaker_wavs:
                speaker_wav_bytes = speaker_wav.read()
                # Convert the uploaded audio file to a WAV format using pydub
                try:
                    audio = AudioSegment.from_file(io.BytesIO(speaker_wav_bytes))
                    wav_buffer = io.BytesIO()
                    audio.export(wav_buffer, format="wav")
                    wav_buffer.seek(0)  # Reset buffer position to the beginning
                except Exception as e:
                    return f"Error processing audio file: {e}"

                temp_wav_file = tempfile.NamedTemporaryFile(suffix=".wav", delete=False)
                temp_wav_file.write(wav_buffer.read())
                temp_wav_file.close()
                temp_files.append(temp_wav_file.name)

        output_buffer = io.BytesIO()
        if temp_files:
            models['multi'].tts_to_file(
                text=text,
                speaker_wav=temp_files, 
                language=language,
                file_path=output_buffer, 
                temperature=temperature,
                length_penalty=length_penalty,
                repetition_penalty=repetition_penalty,
                top_k=top_k,
                top_p=top_p,
                speed=speed,
                enable_text_splitting=enable_text_splitting
            )
        else:
            models['multi'].tts_to_file(
                text=text,
                speaker=speaker_idx, 
                language=language,
                file_path=output_buffer, 
                temperature=temperature,
                length_penalty=length_penalty,
                repetition_penalty=repetition_penalty,
                top_k=top_k,
                top_p=top_p,
                speed=speed,
                enable_text_splitting=enable_text_splitting
            )
        
        output_buffer.seek(0)
        return output_buffer.read()

    finally:
        for temp_file in temp_files:
            if isinstance(temp_file, str) and os.path.exists(temp_file):
                os.remove(temp_file)


# 创建Gradio界面
inputs = [
    gr.Textbox(value="Hello, World!", label="Text to Synthesize"),
    gr.File(file_types=["audio"], label="Speaker WAV files (optional)", optional=True, multiple=True),
    gr.Textbox(value="Ana Florence", label="Speaker Index"),
    gr.Textbox(value="ja", label="Language"),
    gr.Slider(0, 1, value=0.65, step=0.01, label="Temperature"),
    gr.Slider(0.5, 2, value=1.0, step=0.1, label="Length Penalty"),
    gr.Slider(1, 10, value=2.0, step=0.1, label="Repetition Penalty"),
    gr.Slider(1, 100, value=50, step=1, label="Top-K"),
    gr.Slider(0, 1, value=0.8, step=0.01, label="Top-P"),
    gr.Slider(0.5, 2, value=1.0, step=0.01, label="Speed"),
    gr.Checkbox(value=True, label="Enable Text Splitting")
]

outputs = gr.Audio(label="Generated Speech")

gr.Interface(
    fn=synthesize_tts, 
    inputs=inputs, 
    outputs=outputs, 
    title="Text-to-Speech Synthesis with Gradio"
).launch()