app.py

import gradio as gr
import os
import random
import IPython.display as ipd
import commons
import utils
import json
import torch
import tempfile
import numpy as np
from models import SynthesizerTrn
from text.symbols import symbols
from text import text_to_sequence
from scipy.io.wavfile import write

def get_text(text, hps):
    text_norm = text_to_sequence(text, hps.data.text_cleaners)
    if hps.data.add_blank:
        text_norm = commons.intersperse(text_norm, 0)
    text_norm = torch.LongTensor(text_norm)
    return text_norm

def get_text_byroma(text, hps):
    text_norm = []
    for i in text:
        text_norm.append(symbols.index(i))
    if hps.data.add_blank:
        text_norm = commons.intersperse(text_norm, 0)
    text_norm = torch.LongTensor(text_norm)
    return text_norm

hps = utils.get_hparams_from_file("./configs/leo.json")
net_g = SynthesizerTrn(
    len(symbols),
    hps.data.filter_length // 2 + 1,
    hps.train.segment_size // hps.data.hop_length,
    n_speakers=hps.data.n_speakers,
    **hps.model)
_ = net_g.eval()
_ = utils.load_checkpoint("logs/leo/G_4000.pth", net_g, None)

# 随机抽取情感参考音频的根目录
random_emotion_root = "wavs"
emotion_dict = json.load(open("configs/leo.json", "r"))

def tts(txt, emotion, temp_file_path):
    """emotion为参考情感音频路径或random_sample（随机抽取）"""
    if roma:
        stn_tst = get_text_byroma(txt, hps)
    else:
        stn_tst = get_text(txt, hps)
    with torch.no_grad():
        x_tst = stn_tst.unsqueeze(0)
        x_tst_lengths = torch.LongTensor([stn_tst.size(0)])
        sid = torch.LongTensor([0])
        if os.path.exists(f"{emotion}.emo.npy"):
            emo = torch.FloatTensor(np.load(f"{emotion}.emo.npy")).unsqueeze(0)
        elif emotion == "random_sample":
            while True:
                rand_wav = random.sample(os.listdir(random_emotion_root), 1)[0]
                if rand_wav.endswith('wav') and os.path.exists(f"{random_emotion_root}/{rand_wav}.emo.npy"):
                    break
            emo = torch.FloatTensor(np.load(f"{random_emotion_root}/{rand_wav}.emo.npy")).unsqueeze(0)
            print(f"{random_emotion_root}/{rand_wav}")
        elif emotion.endswith("wav"):
            import emotion_extract
            emo = torch.FloatTensor(emotion_extract.extract_wav(emotion))
        else:
            print("emotion参数不正确")

        audio = net_g.infer(x_tst, x_tst_lengths, sid=sid, noise_scale=0.667, noise_scale_w=0.8, length_scale=1.2, emo=emo)[0][0,0].data.float().numpy()

        # Save the numpy array as a temporary file
        write(temp_file_path, hps.data.sampling_rate, audio)

        # Display the audio
        ipd.display(ipd.Audio(temp_file_path, rate=hps.data.sampling_rate, normalize=False))

        # Delete the temporary file
        os.remove(temp_file_path)

    return audio

def generate_audio(txt, emotion):
    # Create a temporary file
    temp_file = tempfile.NamedTemporaryFile(suffix=".wav", delete=False)
    temp_file_path = temp_file.name
    audio = tts(txt, emotion, temp_file_path)
    return audio

input_text = gr.inputs.Textbox(label="输入文本")
input_emotion = gr.inputs.Dropdown(choices=["random_sample"] + os.listdir(random_emotion_root), label="参考情感音频")
output_audio = gr.outputs.Audio(type="numpy", label="合成音频")

iface = gr.Interface(fn=generate_audio, inputs=[input_text, input_emotion], outputs=output_audio)
iface.launch()