Hugging Face's logo

BarryFutureman
/
NeuTTS-Express

Text-to-Speech
Transformers
Safetensors
English
qwen2
text-generation
text-generation-inference
Model card Files
xet
Community
NeuTTS-Express
File size: 2,153 Bytes 
dafc205
 
 
 
 
 
 
 
a1dc315
 
 
 
335551e
 
a1dc315
 
 
 
 
335551e
 
 
3b711a6
 
 
335551e
3b711a6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
335551e
 
 
 
 
 
 
2bdbee6
 
 
 
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
 
---
license: apache-2.0
language:
- en
base_model:
- neuphonic/neutts-air
pipeline_tag: text-to-speech
library_name: transformers
---

# NeuTTS Express

<audio controls src="https://cdn-uploads.huggingface.co/production/uploads/6599dc66eabe0f3e98de7cf6/EyqWd8BtQT-o5nyb6CavE.wav"></audio>

- Compact model size (0.5B)
- Emotional and paralinguistic controls via \(tags\)
- Supports streaming for low latency usecases
- Instant voice cloning

Supported tags: `(neutral)`, `(happy)`, `(sad)`, `(angry)`, `(surprised)`, `(disgusted)`, `(fearful)`, `(giggle)`, `(gasp)`, `(exhale)`, `(laugh)`, `(chuckle)`, `(sign)`

```python
import time
import re
import types
import soundfile as sf
from neuttsair.neutts import NeuTTSAir

# patch function
def patched_to_phones(self, text: str) -> str:
    pattern = r'\([^)]*\)'
    preserved_matches = list(re.finditer(pattern, text))
    
    if not preserved_matches:
        phones = self.phonemizer.phonemize([text])[0]
        return " ".join(phones.split())
    
    parts = re.split(pattern, text)
    preserved_texts = [m.group() for m in preserved_matches]

    phonemized_parts = []
    for part in parts:
        if part.strip():
            phonemized = self.phonemizer.phonemize([part])[0]
            phonemized = " ".join(phonemized.split())
            phonemized_parts.append(phonemized)
        else:
            phonemized_parts.append("")

    result = []
    for i, phonemized_part in enumerate(phonemized_parts):
        if phonemized_part:
            result.append(phonemized_part)
        if i < len(preserved_texts):
            result.append(preserved_texts[i])
    
    return " ".join(result)

# apply monkey patch
NeuTTSAir._to_phones = patched_to_phones

# inference
tts = NeuTTSAir(
    backbone_repo="BarryFutureman/NeuTTS-Express",
    backbone_device="cuda",
    codec_repo="neuphonic/distill-neucodec",
    codec_device="cuda"
)

input_text = "I just got the best news ever (giggle), and I couldn't be happier!"
start_time = time.time()
wav = tts.infer(input_text)
end_time = time.time()

sf.write("output.wav", wav, 24000)
```

Higher `temperature`, `top-p`, `top-k` is recommended.