README.md · VieNeu-TTS-v3-Turbo

1

---

2

license: apache-2.0

3

datasets:

4

- pnnbao-ump/VieNeu-TTS-10k-ENVI

5

language:

6

- vi

7

- en

8

pipeline_tag: text-to-speech

9

tags:

10

- voice-cloning

11

- code-switching

12

- podcast

13

- emotion-control

14

- 48khz

15

---

16

17

# 🦜 VieNeu-TTS v3 Turbo

18

19

[![GitHub](https://img.shields.io/badge/GitHub-Repository-blue)](https://github.com/pnnbao97/VieNeu-TTS)

20

[![Model](https://img.shields.io/badge/Hugging%20Face-Model-yellow)](https://huggingface.co/pnnbao-ump/VieNeu-TTS-v3-Turbo)

21

[![Discord](https://img.shields.io/badge/Discord-Join%20Us-5865F2?logo=discord&logoColor=white)](https://discord.gg/yJt8kzjzWZ)

22

23

## Overview

24

25

<video controls src="https://cdn-uploads.huggingface.co/production/uploads/68b923a86c86c127a1975eda/fY6_ONxsTJ02WgNmeFpfL.mp4" width="100%"></video>

26

27

**VieNeu-TTS v3 Turbo** is the next generation of Vietnamese TTS — **48 kHz** high-fidelity speech, **instant voice cloning**, built-in **multi-speaker default voices**, inline **emotion cues**, and seamless **bilingual (En–Vi) code-switching**. It is a pure-PyTorch engine running on both **GPU and CPU**, using the **MOSS-Audio-Tokenizer-Nano** codec.

28

29

> [!NOTE]

30

> **Early access.** v3 Turbo is released for **preview**. It is fast and natural, but some features (notably the emotion cues) are still **experimental**. The full v3 release is coming in the next few weeks.

31

32

> [!IMPORTANT]

33

> **What's new in v3:**

34

> - **48 kHz audio** — a big jump in fidelity over v2 (24 kHz).

35

> - **Built-in default voices** — each default speaker is addressed by a dedicated speaker token + fixed reference, so the voice is **stable and consistent** with no reference clip needed.

36

> - **Emotion / non-verbal cues** *(experimental)* — drop `[cười]`, `[thở dài]`, `[hắng giọng]` straight into your text.

37

> - **Batched generation** — synthesize many chunks at once (batch size up to 32), including a **multi-speaker conversation** mode that batches the whole script regardless of speaker.

38

> - **Instant Voice Cloning** — still clones a voice from just **3–5 seconds** of audio (cloning is available from v3 onward; v1/v2 do not support it).

39

40

## 🏗️ Architecture & Credits

41

42

The **VieNeu-TTS v3 Turbo architecture is an original design by the author, Phạm Nguyễn Ngọc Bảo**, and is **trained from scratch** on ~10,000 hours of English–Vietnamese speech — it is **not** a fine-tune, distillation, or adaptation of any existing TTS model.

43

44

- **Model architecture & training:** designed and trained from scratch by **Phạm Nguyễn Ngọc Bảo** — https://github.com/pnnbao97

45

- **Audio codec:** [MOSS-Audio-Tokenizer-Nano](https://huggingface.co/OpenMOSS-Team/MOSS-Audio-Tokenizer-Nano) (OpenMOSS-Team) — 48 kHz neural audio codec.

46

- **Phonemizer:** [sea-g2p](https://github.com/pnnbao97/sea-g2p) — fast Vietnamese/English grapheme-to-phoneme, also by the author.

47

48

Tác giả: **Phạm Nguyễn Ngọc Bảo**

49

50

## ☕ Support This Project

51

52

Training high-quality TTS models requires significant GPU resources. If you find this model useful, please consider supporting the development:

53

54

[![Buy Me a Coffee](https://img.shields.io/badge/Buy%20Me%20a%20Coffee-Support-orange?logo=buy-me-a-coffee)](https://buymeacoffee.com/pnnbao)

55

56

---

57

58

## 🔥 Quick Start (Web UI)

59

```bash

60

git clone https://github.com/pnnbao97/VieNeu-TTS.git

61

cd VieNeu-TTS

62

```

63

64

- **Option 1: CPU** (minimal, torch-free) — runs **v3 Turbo via ONNX**

65

66

```bash

67

uv sync

68

```

69

- **Option 2: GPU** — **v3 Turbo (PyTorch) + VieNeu-TTS v2 (GPU)**

70

71

```bash

72

uv sync --group gpu

73

```

74

**Start the Web UI:**

75

76

```bash

77

uv run vieneu-web

78

```

79

80

In the Web UI, pick **"VieNeu-TTS-v3-Turbo (Thử nghiệm)"** as the backbone. You get a **Default voice** tab, a **Voice Cloning** tab, and a **Conversation** tab (batched multi-speaker podcasts).

81

82

---

83

84

## 📦 Using Python SDK (vieneu)

85

86

```bash

87

pip install vieneu

88

```

89

90

### Full Features Guide

91

```python

92

from vieneu import Vieneu

93

from time import time

94

95

# Default = v3 Turbo. CPU → ONNX (torch-free); GPU → PyTorch (auto-detected).

96

tts = Vieneu()

97

98

text = f"""[cười] Trời ơi, cái giọng nó tự nhiên mà nó mượt mà dã man, nghe không khác gì người thật luôn. Giờ thì tha hồ mà quẩy content với cả kho giọng nói đa dạng, đủ mọi sắc thái biểu cảm. Mọi người bật loa lên rồi cùng trải nghiệm thử với mình nhé!"""

99

100

start_time = time()

101

# 1. Default voice (Bình An) — 48 kHz, no reference needed

102

audio = tts.infer(text)

103

tts.save(audio, "output.wav")

104

end_time = time()

105

print(f"Time taken: {end_time - start_time} seconds")

106

# 2. Built-in voices by name

107

for label, voice_id in tts.list_preset_voices():

108

print(label, voice_id)

109

audio = tts.infer("Mình là Xuân Vĩnh nè!", voice="Xuân Vĩnh")

110

tts.save(audio, "output_Xuân Vĩnh.wav")

111

# # 3. Emotion / non-verbal cues — EXPERIMENTAL: [cười] [thở dài] [hắng giọng]

112

# audio = tts.infer("Nghe hay quá đi [cười]. Để mình nói tiếp [hắng giọng].", voice="Ngọc Linh")

113

114

# # 4. Instant voice cloning from a 3–5s reference clip

115

# audio = tts.infer("Đây là giọng được nhân bản tức thì.", ref_audio="my_voice.wav")

116

```

117

118

> [!TIP]

119

> A **temperature around 0.8** gives the most stable result for v3 Turbo. Higher values add expressiveness but can be less stable.

120

121

---

122

123

## 🎭 Default Voices

124

125

Built-in voices — call them by name via `voice="<name>"`, no reference audio required.

126

127

| Voice | Gender | Style |

128

|---|---|---|

129

| **Ngọc Lan** *(default)* | Female | Soft / gentle |

130

| Ngọc Linh | Female | Bright |

131

| Trúc Ly | Female | Youthful |

132

| Mỹ Duyên | Female | Smooth |

133

| Xuân Vĩnh | Male | Upbeat |

134

| Thái Sơn | Male | Firm |

135

| Gia Bảo | Male | Smooth |

136

| Đức Trí | Male | Clear |

137

| Trọng Hữu | Male | Knowledgeable |

138

| Bình An | Male | Even / calm |

139

140

For any other voice, use **Voice Cloning** with a short reference clip (`ref_audio="..."`).

141

142

---

143

144

## 🔬 Model Variants

145

146

147

| --------------------- | ------- | ------- | ----------- | ---------- | ----------------------------------------- |

148

| **VieNeu-TTS-v3-Turbo** | PyTorch | GPU/CPU | 48 kHz      | ⭐⭐⭐⭐⭐ | **Default voices, Cloning, Emotion cues** |

149

| VieNeu-TTS-v2         | PyTorch | GPU/CPU | 24 kHz      | ⭐⭐⭐⭐⭐ | Podcast, En-Vi code-switching             |

150

| VieNeu-TTS-v2 (GGUF) | GGUF Q4 | CPU | 24 kHz | ⭐⭐⭐⭐ | Fastest on CPU, Podcast |

151

| VieNeu-TTS-v1 | PyTorch | GPU | 24 kHz | ⭐⭐⭐⭐ | Stable (Vi only) |

152

153

---

154

155

## License

156

157

This model package is distributed under [Apache License 2.0](https://www.apache.org/licenses/LICENSE-2.0), matching the upstream model repository.

158

159

When you reuse, redistribute, or convert these assets, please keep the license notice and attribution intact for both:

160

161

- the original project: [pnnbao97/VieNeu-TTS](https://github.com/pnnbao97/VieNeu-TTS)

162

- this Hugging Face package: [pnnbao-ump/VieNeu-TTS-v3-Turbo](https://huggingface.co/pnnbao-ump/VieNeu-TTS-v3-Turbo)

163

164

If you bundle additional third-party assets, their own licenses still apply as well.

165

166

---

167

## 📑 Citation

168

169

```bibtex

170

@misc{vieneutts2026,

171

title = {VieNeu-TTS v3 Turbo: 48kHz Vietnamese Text-to-Speech with Instant Voice Cloning and Emotion Control},

172

author = {Pham Nguyen Ngoc Bao},

173

year = {2026},

174

publisher = {Hugging Face},

175

howpublished = {\url{https://huggingface.co/pnnbao-ump/VieNeu-TTS-v3-Turbo}}

176

}

177

```

178

179

---

180

181

**Made with ❤️ for the Vietnamese TTS community**

182