File size: 5,200 Bytes
5cc336f
 
 
 
 
 
 
923f534
5cc336f
 
0dfa563
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
e042373
 
0dfa563
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
---
license: mit
datasets:
- OpenSpeechHub/mls_eng_10k_snac_qwen
language:
- en
base_model:
- Qwen/Qwen3-0.6B
pipeline_tag: text-to-speech
library_name: transformers
---
## Overview
VyvoTTS-v0-Qwen3-0.6B is a Text-to-Speech model based on Qwen3-0.6B, trained to produce natural-sounding English speech.

- **Type:** Text-to-Speech  
- **Language:** English  
- **License:** MIT  
- **Params:** ~810M  

> **Note:** This model has a high Word Error Rate (WER) as it was trained on a 10,000-hour dataset. To improve the model's accuracy, you should use it as a pretrained base.
> I can recommend the Emilia dataset for this purpose. After the pretraining process is complete, you should perform fine-tuning for single-speaker speech.

## Usage
Below is an example of using the model with `unsloth` and `SNAC` for speech generation:

```python
from unsloth import FastLanguageModel
import torch
from snac import SNAC

model, tokenizer = FastLanguageModel.from_pretrained(
    model_name = "Vyvo/VyvoTTS-v0-Qwen3-0.6B",
    max_seq_length= 8192,
    dtype = None,
    load_in_4bit = False,
)
snac_model = SNAC.from_pretrained("hubertsiuzdak/snac_24khz")
tokeniser_length = 151669
start_of_text = 151643
end_of_text = 151645

start_of_speech = tokeniser_length + 1
end_of_speech = tokeniser_length + 2
start_of_human = tokeniser_length + 3
end_of_human = tokeniser_length + 4
pad_token = tokeniser_length + 7

audio_tokens_start = tokeniser_length + 10
prompts = ["Hey there my name is Elise, and I'm a speech generation model that can sound like a person."]
chosen_voice = None

FastLanguageModel.for_inference(model)
snac_model.to("cpu")
prompts_ = [(f"{chosen_voice}: " + p) if chosen_voice else p for p in prompts]

all_input_ids = []
for prompt in prompts_:
    input_ids = tokenizer(prompt, return_tensors="pt").input_ids
    all_input_ids.append(input_ids)

start_token = torch.tensor([[start_of_human]], dtype=torch.int64)
end_tokens = torch.tensor([[end_of_text, end_of_human]], dtype=torch.int64)

all_modified_input_ids = []
for input_ids in all_input_ids:
    modified_input_ids = torch.cat([start_token, input_ids, end_tokens], dim=1)
    all_modified_input_ids.append(modified_input_ids)

all_padded_tensors, all_attention_masks = [], []
max_length = max([m.shape[1] for m in all_modified_input_ids])
for m in all_modified_input_ids:
    padding = max_length - m.shape[1]
    padded_tensor = torch.cat([torch.full((1, padding), pad_token, dtype=torch.int64), m], dim=1)
    attention_mask = torch.cat([torch.zeros((1, padding), dtype=torch.int64), torch.ones((1, m.shape[1]), dtype=torch.int64)], dim=1)
    all_padded_tensors.append(padded_tensor)
    all_attention_masks.append(attention_mask)

input_ids = torch.cat(all_padded_tensors, dim=0).to("cuda")
attention_mask = torch.cat(all_attention_masks, dim=0).to("cuda")

generated_ids = model.generate(
    input_ids=input_ids,
    attention_mask=attention_mask,
    max_new_tokens=1200,
    do_sample=True,
    temperature=0.6,
    top_p=0.95,
    repetition_penalty=1.1,
    num_return_sequences=1,
    eos_token_id=end_of_speech,
    use_cache=True
)

token_to_find = start_of_speech
token_to_remove = end_of_speech
token_indices = (generated_ids == token_to_find).nonzero(as_tuple=True)

if len(token_indices[1]) > 0:
    last_occurrence_idx = token_indices[1][-1].item()
    cropped_tensor = generated_ids[:, last_occurrence_idx+1:]
else:
    cropped_tensor = generated_ids

processed_rows = []
for row in cropped_tensor:
    masked_row = row[row != token_to_remove]
    processed_rows.append(masked_row)

code_lists = []
for row in processed_rows:
    row_length = row.size(0)
    new_length = (row_length // 7) * 7
    trimmed_row = row[:new_length]
    trimmed_row = [t - audio_tokens_start for t in trimmed_row]
    code_lists.append(trimmed_row)

def redistribute_codes(code_list):
    layer_1, layer_2, layer_3 = [], [], []
    for i in range((len(code_list)+1)//7):
        layer_1.append(code_list[7*i])
        layer_2.append(code_list[7*i+1]-4096)
        layer_3.append(code_list[7*i+2]-(2*4096))
        layer_3.append(code_list[7*i+3]-(3*4096))
        layer_2.append(code_list[7*i+4]-(4*4096))
        layer_3.append(code_list[7*i+5]-(5*4096))
        layer_3.append(code_list[7*i+6]-(6*4096))
    codes = [
        torch.tensor(layer_1).unsqueeze(0),
        torch.tensor(layer_2).unsqueeze(0),
        torch.tensor(layer_3).unsqueeze(0)
    ]
    audio_hat = snac_model.decode(codes)
    return audio_hat

my_samples = []
for code_list in code_lists:
    samples = redistribute_codes(code_list)
    my_samples.append(samples)

from IPython.display import display, Audio
if len(prompts) != len(my_samples):
    raise Exception("Number of prompts and samples do not match")
else:
    for i in range(len(my_samples)):
        print(prompts[i])
        samples = my_samples[i]
        display(Audio(samples.detach().squeeze().to("cpu").numpy(), rate=24000))

del my_samples, samples
```

## Citation

If you use this model, please cite:

```bibtex
@misc{VyvoTTS-v0-Qwen3-0.6B,
  title={VyvoTTS-v0-Qwen3-0.6B},
  author={Vyvo},
  year={2025},
  howpublished={\url{https://huggingface.co/Vyvo/VyvoTTS-v0-Qwen3-0.6B}}
}
```