---
license: apache-2.0
language:
  - en
tags:
  - text-to-speech
  - vits
  - ax650
  - npu
  - onnx
pipeline_tag: text-to-speech
---

# Inflect-Nano-v2 for AX650 NPU

High-quality English TTS (VITS architecture) optimized for **AX650 NPU3**, running entirely on-chip with zero PyTorch dependency at inference time.

- **Encoder + Decoder**: AX650 NPU (AXMODEL, U16 precision)
- **Duration Predictor**: ONNX Runtime (onnxruntime)
- **Embedding**: NumPy lookup
- **Real-time Factor**: ~0.04x (24× faster than real-time)

## 🎧 Audio Samples

*All samples generated on AX650 NPU, 24kHz.*

**"Hello and welcome to Inflect Nano V2 text to speech."** (3.3s, RTF 0.06x)
<audio controls src="https://huggingface.co/AXERA-TECH/inflect_nano_v2/resolve/main/samples/en_hello.wav"></audio>

**"The weather today is sunny with a gentle breeze."** (2.4s, RTF 0.07x)
<audio controls src="https://huggingface.co/AXERA-TECH/inflect_nano_v2/resolve/main/samples/en_weather.wav"></audio>

**"Once upon a time, there lived a curious robot who loved to sing."** (3.7s, RTF 0.05x)
<audio controls src="https://huggingface.co/AXERA-TECH/inflect_nano_v2/resolve/main/samples/en_story.wav"></audio>

**"Neural voice synthesis is the next frontier of artificial intelligence."** (3.9s, RTF 0.05x)
<audio controls src="https://huggingface.co/AXERA-TECH/inflect_nano_v2/resolve/main/samples/en_tech.wav"></audio>

**"This audio was generated entirely on the AX650 NPU chip."** (4.4s, RTF 0.04x)
<audio controls src="https://huggingface.co/AXERA-TECH/inflect_nano_v2/resolve/main/samples/en_intro.wav"></audio>

## 🚀 Quick Start

```bash
bash setup.sh
bash run.sh "Hello world." output.wav
```

## 📡 API

```bash
# CLI
python -m inflect_tts_sdk --text "Hello" --output out.wav

# OpenAI-compatible Server
python -m inflect_tts_sdk.server --port 8000
```

## 📊 Performance

| Component | Time | Device |
|-----------|------|--------|
| Encoder | 2.4 ms | NPU |
| Decoder | 89.6 ms | NPU |
| DP + Align | 14.7 ms | CPU |
| **Total RTF** | **~0.04x** | 24× real-time |

*Measured on AX650 NPU3, 105 tokens → 2.7s audio.*

## 🧩 Model

| Component | Size | Input | Output |
|-----------|------|-------|--------|
| Encoder | 1.0 MB | x_emb [1,72,200] | m_p, logs_p, x, x_mask |
| Decoder | 5.6 MB | z_p [1,128,500] | waveform [1,1,128000] |

- Quantization: Encoder Conv U16 / MatMul INT8, Decoder Conv+ConvTranspose U16
- Parameters: 4.0M
- Sample rate: 24 kHz
- Source: [owensong/Inflect-Nano-v2](https://hf-mirror.com/owensong/Inflect-Nano-v2)
- Compiled: Pulsar2 6.0-lite, AX650 NPU3

## 📁 File Structure

- `models/` — AXMODEL, dp.onnx, emb_weight.npy
- `python/inflect_tts_sdk/` — Python SDK (CLI + OpenAI Server)
- `cpp/inflect_tts_sdk/` — C++ SDK (CLI + Server)
- `samples/` — audio samples

## License & Attribution（许可与署名）

本仓库是 [owensong/Inflect-Nano-v2](https://huggingface.co/owensong/Inflect-Nano-v2)
（作者 Owen Song，Apache-2.0）的 NPU 转换与再分发。原始 Inflect 代码与权重以
Apache License 2.0 发布；本仓库的转换产物（AXMODEL、SDK）同样以 Apache-2.0 许可提供。

- 许可全文：`LICENSE`（Apache License 2.0）
- 署名与修改说明：`NOTICE.md`
- 第三方组件声明：`THIRD_PARTY_NOTICES.md`、`third_party/`