CosyVoice/webui.py

# Copyright (c) 2024 Alibaba Inc (authors: Xiang Lyu, Liu Yue)
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#   http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import os
import sys
ROOT_DIR = os.path.dirname(os.path.abspath(__file__))
sys.path.append('{}/third_party/Matcha-TTS'.format(ROOT_DIR))

import argparse
import gradio as gr
import numpy as np
import torch
import torchaudio
import random
import librosa

import logging
logging.getLogger('matplotlib').setLevel(logging.WARNING)

from cosyvoice.cli.cosyvoice import CosyVoice
from cosyvoice.utils.file_utils import load_wav, speed_change

logging.basicConfig(level=logging.DEBUG,
                    format='%(asctime)s %(levelname)s %(message)s')

def generate_seed():
    seed = random.randint(1, 100000000)
    return {
        "__type__": "update",
        "value": seed
    }

def set_all_random_seed(seed):
    random.seed(seed)
    np.random.seed(seed)
    torch.manual_seed(seed)
    torch.cuda.manual_seed_all(seed)

max_val = 0.8
def postprocess(speech, top_db=60, hop_length=220, win_length=440):
    speech, _ = librosa.effects.trim(
        speech, top_db=top_db,
        frame_length=win_length,
        hop_length=hop_length
    )
    if speech.abs().max() > max_val:
        speech = speech / speech.abs().max() * max_val
    speech = torch.concat([speech, torch.zeros(1, int(target_sr * 0.2))], dim=1)
    return speech

inference_mode_list = ['预训练音色', '3s极速复刻', '跨语种复刻', '自然语言控制']
instruct_dict = {'预训练音色': '1. 选择预训练音色\n2. 点击生成音频按钮',
                 '3s极速复刻': '1. 选择prompt音频文件，或录入prompt音频，注意不超过30s，若同时提供，优先选择prompt音频文件\n2. 输入prompt文本\n3. 点击生成音频按钮',
                 '跨语种复刻': '1. 选择prompt音频文件，或录入prompt音频，注意不超过30s，若同时提供，优先选择prompt音频文件\n2. 点击生成音频按钮',
                 '自然语言控制': '1. 选择预训练音色\n2. 输入instruct文本\n3. 点击生成音频按钮'}
def change_instruction(mode_checkbox_group):
    return instruct_dict[mode_checkbox_group]

def generate_audio(tts_text, mode_checkbox_group, sft_dropdown, prompt_text, prompt_wav_upload, prompt_wav_record, instruct_text, seed, speed_factor):
    if prompt_wav_upload is not None:
        prompt_wav = prompt_wav_upload
    elif prompt_wav_record is not None:
        prompt_wav = prompt_wav_record
    else:
        prompt_wav = None
    # if instruct mode, please make sure that model is iic/CosyVoice-300M-Instruct and not cross_lingual mode
    if mode_checkbox_group in ['自然语言控制']:
        if cosyvoice.frontend.instruct is False:
            gr.Warning('您正在使用自然语言控制模式, {}模型不支持此模式, 请使用iic/CosyVoice-300M-Instruct模型'.format(args.model_dir))
            return (target_sr, default_data)
        if instruct_text == '':
            gr.Warning('您正在使用自然语言控制模式, 请输入instruct文本')
            return (target_sr, default_data)
        if prompt_wav is not None or prompt_text != '':
            gr.Info('您正在使用自然语言控制模式, prompt音频/prompt文本会被忽略')
    # if cross_lingual mode, please make sure that model is iic/CosyVoice-300M and tts_text prompt_text are different language
    if mode_checkbox_group in ['跨语种复刻']:
        if cosyvoice.frontend.instruct is True:
            gr.Warning('您正在使用跨语种复刻模式, {}模型不支持此模式, 请使用iic/CosyVoice-300M模型'.format(args.model_dir))
            return (target_sr, default_data)
        if instruct_text != '':
            gr.Info('您正在使用跨语种复刻模式, instruct文本会被忽略')
        if prompt_wav is None:
            gr.Warning('您正在使用跨语种复刻模式, 请提供prompt音频')
            return (target_sr, default_data)
        gr.Info('您正在使用跨语种复刻模式, 请确保合成文本和prompt文本为不同语言')
    # if in zero_shot cross_lingual, please make sure that prompt_text and prompt_wav meets requirements
    if mode_checkbox_group in ['3s极速复刻', '跨语种复刻']:
        if prompt_wav is None:
            gr.Warning('prompt音频为空，您是否忘记输入prompt音频？')
            return (target_sr, default_data)
        if torchaudio.info(prompt_wav).sample_rate < prompt_sr:
            gr.Warning('prompt音频采样率{}低于{}'.format(torchaudio.info(prompt_wav).sample_rate, prompt_sr))
            return (target_sr, default_data)
    # sft mode only use sft_dropdown
    if mode_checkbox_group in ['预训练音色']:
        if instruct_text != '' or prompt_wav is not None or prompt_text != '':
            gr.Info('您正在使用预训练音色模式，prompt文本/prompt音频/instruct文本会被忽略！')
    # zero_shot mode only use prompt_wav prompt text
    if mode_checkbox_group in ['3s极速复刻']:
        if prompt_text == '':
            gr.Warning('prompt文本为空，您是否忘记输入prompt文本？')
            return (target_sr, default_data)
        if instruct_text != '':
            gr.Info('您正在使用3s极速复刻模式，预训练音色/instruct文本会被忽略！')

    if mode_checkbox_group == '预训练音色':
        logging.info('get sft inference request')
        set_all_random_seed(seed)
        output = cosyvoice.inference_sft(tts_text, sft_dropdown)
    elif mode_checkbox_group == '3s极速复刻':
        logging.info('get zero_shot inference request')
        prompt_speech_16k = postprocess(load_wav(prompt_wav, prompt_sr))
        set_all_random_seed(seed)
        output = cosyvoice.inference_zero_shot(tts_text, prompt_text, prompt_speech_16k)
    elif mode_checkbox_group == '跨语种复刻':
        logging.info('get cross_lingual inference request')
        prompt_speech_16k = postprocess(load_wav(prompt_wav, prompt_sr))
        set_all_random_seed(seed)
        output = cosyvoice.inference_cross_lingual(tts_text, prompt_speech_16k)
    else:
        logging.info('get instruct inference request')
        set_all_random_seed(seed)
        output = cosyvoice.inference_instruct(tts_text, sft_dropdown, instruct_text)

    if speed_factor != 1.0:
        try:
            audio_data, sample_rate = speed_change(output["tts_speech"], target_sr, str(speed_factor))
            audio_data = audio_data.numpy().flatten()
        except Exception as e:
            print(f"Failed to change speed of audio: \n{e}")
    else:
        audio_data = output['tts_speech'].numpy().flatten()

    return (target_sr, audio_data)

def main():
    with gr.Blocks() as demo:
        gr.Markdown("### 代码库 [CosyVoice](https://github.com/FunAudioLLM/CosyVoice) 预训练模型 [CosyVoice-300M](https://www.modelscope.cn/models/iic/CosyVoice-300M) [CosyVoice-300M-Instruct](https://www.modelscope.cn/models/iic/CosyVoice-300M-Instruct) [CosyVoice-300M-SFT](https://www.modelscope.cn/models/iic/CosyVoice-300M-SFT)")
        gr.Markdown("#### 请输入需要合成的文本，选择推理模式，并按照提示步骤进行操作")

        tts_text = gr.Textbox(label="输入合成文本", lines=1, value="我是通义实验室语音团队全新推出的生成式语音大模型，提供舒适自然的语音合成能力。")
        speed_factor = gr.Slider(minimum=0.25, maximum=4, step=0.05, label="语速调节", value=1.0, interactive=True)
        with gr.Row():
            mode_checkbox_group = gr.Radio(choices=inference_mode_list, label='选择推理模式', value=inference_mode_list[0])
            instruction_text = gr.Text(label="操作步骤", value=instruct_dict[inference_mode_list[0]], scale=0.5)
            sft_dropdown = gr.Dropdown(choices=sft_spk, label='选择预训练音色', value=sft_spk[0], scale=0.25)
            with gr.Column(scale=0.25):
                seed_button = gr.Button(value="\U0001F3B2")
                seed = gr.Number(value=0, label="随机推理种子")

        with gr.Row():
            prompt_wav_upload = gr.Audio(sources='upload', type='filepath', label='选择prompt音频文件，注意采样率不低于16khz')
            prompt_wav_record = gr.Audio(sources='microphone', type='filepath', label='录制prompt音频文件')
        prompt_text = gr.Textbox(label="输入prompt文本", lines=1, placeholder="请输入prompt文本，需与prompt音频内容一致，暂时不支持自动识别...", value='')
        instruct_text = gr.Textbox(label="输入instruct文本", lines=1, placeholder="请输入instruct文本.", value='')

        generate_button = gr.Button("生成音频")

        audio_output = gr.Audio(label="合成音频")

        seed_button.click(generate_seed, inputs=[], outputs=seed)
        generate_button.click(generate_audio,
                              inputs=[tts_text, mode_checkbox_group, sft_dropdown, prompt_text, prompt_wav_upload, prompt_wav_record, instruct_text, seed, speed_factor],
                              outputs=[audio_output])
        mode_checkbox_group.change(fn=change_instruction, inputs=[mode_checkbox_group], outputs=[instruction_text])
    demo.queue(max_size=4, default_concurrency_limit=2)
    demo.launch(server_name='0.0.0.0', server_port=args.port)

if __name__ == '__main__':
    parser = argparse.ArgumentParser()
    parser.add_argument('--port',
                        type=int,
                        default=8000)
    parser.add_argument('--model_dir',
                        type=str,
                        default='iic/CosyVoice-300M',
                        help='local path or modelscope repo id')
    args = parser.parse_args()
    cosyvoice = CosyVoice(args.model_dir)
    sft_spk = cosyvoice.list_avaliable_spks()
    prompt_sr, target_sr = 16000, 22050
    default_data = np.zeros(target_sr)
    main()