Ai生成语音怎么调速(paddleSpeech)

依赖源码finetuneTTS.py 的功能，搜索这个项目就能找到这个文件，把这些文件拷到本地使用，以下是核心的两步

1.生成音素

2.生成语音

生成音素.py

import os
from Text2Speech.util.finetuneTTS import load_fs2_model, fs2_inference, get_tts_phone_ids, \
    get_idx2ph_dict, list_to_durations, \
    duration_phones_to_list_fix_speed
import paddle
import json


def generate_yinsu(text, gen_exp_name, outPath, speed=1):
    ff = open(text, "r", encoding="utf-8")
    sentences = ff.read()
    ff.close()
    text_list = sentences.split("\n")

    model = load_fs2_model(gen_exp_name)


    num = 0
    for i in text_list:
        phone_ids = get_tts_phone_ids(i, gen_exp_name, exp_fun=True)
        duration = fs2_inference(model, text=phone_ids[0], spk_id=paddle.to_tensor(0), alpha=1.0, duration=None,
                                 return_duration=True)
        idx2ph_dict = get_idx2ph_dict(gen_exp_name)
        phs_lists = duration_phones_to_list_fix_speed(duration, phone_ids, idx2ph_dict,speed)
        phs_lists_json = outPath + "/phs_list.json" + f'{num}'
        if os.path.exists(phs_lists_json):
            os.remove(phs_lists_json)
        with open(phs_lists_json, "w", encoding="utf8") as f:
            json.dump(phs_lists, f, indent=3)
        num += 1
    print("------------完成因素生成--------------")

def read_yinsu(outPath,index):
    phs_lists_json = outPath + "/phs_list.json" + f'{index}'
    if os.path.exists(phs_lists_json):
        with open(phs_lists_json, "r", encoding="utf8") as f:
            phs_lists = json.load(f)
    new_duration = list_to_durations(phs_lists)
    print(new_duration)
    return new_duration

if __name__ == '__main__':
    rootPath = "D:/Text2Speech/"
    # 模型路径
    am_inference_dir = rootPath + "model/" + "guanbo2"
    # 声码器路径
    voc_inference_dir = rootPath + "soundCoder/" + "pwgan_aishell3_static_1.1.0"
    # 合成音频输出路径
    wav_output_dir = rootPath + "output"
    # 加速设备cpu gpu,amd pytorch-dml仍然很多不兼容，使用cpu就好
    device = "cpu"
    # 想要生成语音的文本文档对应文件名
    txt_name = rootPath + "speechtxt.txt"

    # 生成音素可以对特定词语 语句进行微调，应该先生成这个数据，后面使用这个数据，有什么不对的可以进行调整
    generate_yinsu(txt_name, am_inference_dir, wav_output_dir,-1.4)