utils/create_audiobook.py

import torch
from TTS.api import TTS
import os
import numpy as np
from scipy.io import wavfile

# Save the original torch.load function
_original_torch_load = torch.load

# Define a custom function that forces weights_only=False
def custom_torch_load(*args, **kwargs):
    if "weights_only" not in kwargs:
        kwargs["weights_only"] = False
    return _original_torch_load(*args, **kwargs)

# Override torch.load globally
torch.load = custom_torch_load

device = "cuda" if torch.cuda.is_available() else "cpu"
print(f"Using device: {device}")


# generate_samples()
def create_audiobook(gender_list,english_transcript,speaker_list):
    parent_wav = []
    tts = TTS("tts_models/multilingual/multi-dataset/xtts_v2").to(device)
    for text in english_transcript:
        print(text['char_id'],text['text'])
        voice_gender = 'Male'
        voice_speaker = ''
        if text['char_id'] != None:
            voice_gender = gender_list[int(text['char_id'])][0]
            print("voice person",speaker_list[text['char_id']])
            voice_speaker = speaker_list[text['char_id']]
        else:
            print("voice person",speaker_list[-1])
            voice_speaker = speaker_list[-1]
        print("Voice Gender",voice_gender)

        try:

            wav = tts.tts(
                text=text['text'],
                speaker=voice_speaker.replace('_',' '),
                language="en"
            )
            print("✓ Audio Generated")
            parent_wav.append(wav)

        except Exception as e:
            print(f"✗ Error with : {e}")
    combined = np.concatenate(parent_wav)
    return combined,tts.synthesizer.output_sample_rate

# parent_wav = generate_audio(gender_list,english_transcript,tts)