Open Source AI Voice: A Developer's Guide to Free TTS

1import torch
2from TTS.api import TTS
3
4# Get device
5device = "cuda" if torch.cuda.is_available() else "cpu"
6
7# Init TTS
8tts = TTS(model_name="tts_models/multilingual/multi-dataset/xtts_v2", progress_bar=False, gpu=True if device == "cuda" else False)
9tts.to(device)
10
11# Run TTS
12tts.tts_to_file(text="This is a test! 🐸", speaker_wav="my/wav/file.wav", language="en", file_path="output.wav")
13

1import os
2from TTS.config import load_config
3from TTS.utils.audio import AudioProcessor
4from TTS.tts.models.vits import Vits
5from TTS.tts.utils.text.tokenizer import TTSTokenizer
6
7# Define training configuration
8config_path = "path/to/config.json"  # Replace with your config file
9c = load_config(config_path)
10ap = AudioProcessor(**c.audio)
11
12# Initialize tokenizer
13tokenizer, config = TTSTokenizer.init_from_config(c)
14
15# Initialize model
16model = Vits(config, ap, tokenizer)
17
18# This is just a stub, the full training loop requires more setup
19# including data loading and optimization.
20# Check the Coqui TTS documentation for detailed examples.
21print("Model initialized. Ready for training!")
22

1from espnet2.bin.tts_inference import Text2Speech
2
3# Load the pre-trained model
4t2s = Text2Speech(
5    train_config="path/to/train_config.yaml",  # Replace with your training config
6    model_file="path/to/model.pth",  # Replace with your model path
7    device="cuda",  # Use "cpu" if no GPU available
8    # Optional arguments
9    # Only necessary for multilingual models
10    # requires_spk_embed=True, 
11    # spk_embed_dim=128,
12)
13
14# Generate speech from text
15with torch.no_grad():
16    wav = t2s("Hello, world!")["wav"]
17
18# Save the generated audio
19sf.write("output.wav", wav.view(-1).cpu().numpy(), samplerate=t2s.fs)
20
21