coqui_tts.py

import os
import noisereduce as nr
import soundfile as sf
# from moviepy.editor import *
import string
import json
from glob import glob
import torchaudio
import subprocess
import shutil
import pyloudnorm as pyln
import torch
from TTS.api import TTS
import string

def remove_punctuation(sentence):
    translator = str.maketrans('', '', string.punctuation)
    sentence = sentence.translate(translator)

    # Remove line breaks
    sentence = sentence.replace('\n', ' ').replace('\r', '')

    return sentence


def run_audio_generation_v1(new_text,accent='None'):
    
    new_text =  new_text.replace('\n', ' ').replace('\r', '')
    new_text_mod = remove_punctuation(new_text)

    new_text_split = new_text_mod.split()
    for word in new_text_split:
        if len(word)>=2 and word.isupper():
            new_text = new_text.replace(word, " ".join([*word]))
            
    models = TTS().list_models()
    with open('models.txt', 'w') as f:
        f.writelines(f"{model}\n" for model in models)
    gpu = True if torch.cuda.is_available() else False
    tts = TTS("tts_models/multilingual/multi-dataset/xtts_v2", gpu=gpu) # gpu should be true when server (cuda)
    # if not gpu:
        
    # pre-process story audio file
    # convert to 16 bit mono
    # remove noise
    speaker_wav_data, speaker_wav_rate = sf.read("./tmp/audio/input_src/0.wav")
    speaker_wav_data_no_noise = nr.reduce_noise(y=speaker_wav_data, sr=speaker_wav_rate)
    sf.write('./tmp/audio/speaker_wav.wav', speaker_wav_data_no_noise, speaker_wav_rate, subtype='PCM_16')

    tts.tts_to_file(
        new_text,
        speaker_wav="./tmp/audio/speaker_wav.wav",
        language="en",
        file_path="./tmp/audio/generated-custom.wav"
    )