Text-to-Speech-TTS-ONNX

Utilizes ONNX Runtime for TTS model.

Features

1. Supported Models:

   • KaniTTS
   • F5-TTS
   • IndexTTS
   • BigVGAN (It is part of the TTS module.)

2. End-to-End Processing:

   • The solution includes internal STFT/ISTFT processing.
   • Input: reference audio + text
   • Output: generated speech

3. Optimize:

   • The key components enable 100% deployment of GPU operators.

4. Resources:

   • Explore More Projects

---

性能 Performance

OS	Device	Backend	Model	Time Cost in Seconds (reference audio: 6s / generates approximately 15 words of speech)	RTF
Ubuntu-24.04	Laptop	CPU i7-1165G7	F5-TTS F32	180 (NFE=32)	60
Ubuntu-24.04	Laptop	GPU MX150	F5-TTS F32	62 (NFE=32)	21
Ubuntu-24.04	Laptop	CPU i7-1165G7	IndexTTS F32	18	6
Ubuntu-24.04	Laptop	GPU MX150	BigVGAN V2 24khz_100band_256x F16	4.6 input mel = (1, 100, 512)	1.53
Ubuntu-24.04	Laptop	CPU i7-1165G7	KaniTTS Q8F32	4.2	1.4
Ubuntu-24.04	Laptop	CPU i7-1165G7	KaniTTS Q4F32	2.6	0.87

---

To-Do List

• Beam Search
• VoxCPM

---

Text-to-Speech-TTS-ONNX

通过 ONNX Runtime 实现运行 TTS 模型。

功能

1. 支持的模型：

   • KaniTTS
   • F5-TTS
   • IndexTTS
   • BigVGAN (它是TTS模块的一部分)

2. 端到端处理：

   • 解决方案内置 STFT/ISTFT 处理。
   • 输入：参考音频 + 文本
   • 输出：生成的语音

3. 优化:

   • 模型关键组件实现了 100% GPU 算子部署。

4. 资源：

   • 探索更多项目

---