PyVoice

A Python-based speech processing tool with ASR (Speech-to-Text) and TTS (Text-to-Speech) capabilities, powered by ONNX Runtime.

---

Features

• 🎙 Speech-to-Text (ASR) - Paraformer model for Chinese/English recognition
• 🔊 Text-to-Speech (TTS) - MeloTTS for natural Chinese/English synthesis
• 🚀 FastAPI Server - RESTful API with Swagger documentation
• 📦 Python SDK - Easy integration into other projects
• 🎵 Multi-format Audio - Supports WAV, MP3, FLAC, OGG (via FFmpeg)
• ⚡ Optimized Performance - NumPy vectorization for 10-100x speedup
• 🖥 CLI Tool - Simple command-line interface

---

Quick Start

Installation

# Clone and setup
git clone <repo-url>
cd PyVoice

# Create virtual environment
python3 -m venv venv
source venv/bin/activate  # macOS/Linux

# Install dependencies
pip install -r requirements.txt

# Install ffmpeg (for MP3/FLAC support)
brew install ffmpeg  # macOS
# or: sudo apt install ffmpeg  # Linux

Download Models

git lfs install
git clone https://huggingface.co/getcharzp/go-speech ./temp_models
mv ./temp_models/melo_weights ./melo_weights
mv ./temp_models/paraformer_weights ./paraformer_weights
rm -rf ./temp_models

---

Usage

1. CLI Tool

# Text-to-Speech
python3 main.py tts "Hello, world!" --out output.wav

# Speech-to-Text
python3 main.py asr audio.wav

2. FastAPI Server

python api.py

Then visit:

• Web UI: http://localhost:8000
• API Docs: http://localhost:8000/docs

API Endpoints:

Method	Endpoint	Description
POST	/asr	Upload audio → Get text
POST	/tts	Send text → Get audio (WAV/MP3)
GET	/records	Get history records
GET	/audio/{id}	Download audio file

3. Python SDK

from pyvoice import ASR, TTS

# Speech Recognition
asr = ASR()
text = asr.recognize("audio.mp3")  # Supports WAV, MP3, FLAC, etc.

# Text-to-Speech
tts = TTS()
tts.synthesize_to_file("Hello world", "output.wav")

---

Project Structure

PyVoice/
├── api.py              # FastAPI server
├── pyvoice.py          # SDK interface
├── main.py             # CLI tool
├── asr/                # ASR engine (Paraformer)
├── tts/                # TTS engine (MeloTTS)
├── internal/           # Audio processing utilities
├── storage/            # API uploads/outputs (auto-created)
├── paraformer_weights/ # ASR model files
└── melo_weights/       # TTS model files

---

Tech Stack

• Runtime: Python 3.8+, ONNX Runtime
• API: FastAPI, Uvicorn
• Audio: NumPy, SciPy, FFmpeg
• NLP: jieba (Chinese segmentation)
• Storage: SQLite

---

Performance

Audio processing optimized with NumPy vectorization:

Operation	Optimization	Speedup
Pre-emphasis	Vectorized array ops	~50x
FFT	Batch np.fft.fft	~20x
Mel filterbank	Matrix multiplication	~100x

---

License

MIT License

---

Acknowledgments

Based on getcharzp/go-speech, refactored into a Python SDK with FastAPI integration.