chatterbox_tts_model_init_and_text_to_speech_generation.py

1import torch
2from chatterbox_tts import Chatterbox
3
4# instantiate the chatterbox model
5
6model = Chatterbox(
7    dim = 512,
8    depth = 8,
9    heads = 8,
10    num_codes = 1024,
11    codebook_dim = 128
12)
13
14# mock text and speech tokens (for demonstration of the interface)
15# in a real scenario, text would be tokenized and audio would be encoded by a codec
16
17text_tokens = torch.randint(0, 100, (1, 128))
18speech_tokens = torch.randint(0, 1024, (1, 256))
19
20# forward pass
21
22loss = model(
23    text_tokens,
24    speech_tokens,
25    return_loss = True
26)
27
28loss.backward()
29
30# after training, generating speech from text
31
32generated_speech = model.generate(text_tokens) # (1, 512) - generated codes