Parrots is an open-source toolkit providing integrated Automatic Speech Recognition (ASR) and Text-To-Speech (TTS) capabilities, designed for ease of use with a focus on Chinese and English. It targets developers and researchers needing a quick way to implement speech processing tasks, offering pre-trained models for multi-lingual and multi-character voice synthesis.

How It Works

The ASR component leverages distilwhisper for robust speech-to-text conversion across multiple languages. The TTS engine is built upon GPT-SoVITS, enabling high-quality voice synthesis with support for various languages and distinct speaker identities, including singing voices. This dual approach allows for a unified, "out-of-the-box" solution for common speech AI applications.

Quick Start & Requirements

• Install via pip: pip install torch (or conda), then pip install parrots.
• Alternatively, clone the repo and run python setup.py install.
• Requires PyTorch.
• Official Demo: https://www.mulanai.com/product/tts/
• HuggingFace Demo: https://huggingface.co/spaces/shibing624/parrots
• Example usage scripts are provided in the examples/ directory.

Highlighted Details

• Supports ASR for Chinese, English, and other languages.
• TTS supports Chinese, English, Japanese, and more, with pre-trained speakers like "MaiMai" (singing female anchor) and "KusanagiNene" (loli).
• Offers both Python API and command-line interface (CLI) for ASR and TTS tasks.
• Models are available on HuggingFace, simplifying integration.

Maintenance & Community

• Project maintained by shibing624.
• Contact via email (xuming624@qq.com) or WeChat (xuming624) for the Python-NLP exchange group.
• Encourages contributions with unit tests.

Licensing & Compatibility

• Licensed under the Apache License 2.0.
• Permissive for commercial use, but requires attribution to Parrots and the license in product documentation.

Limitations & Caveats

The project code is described as "rough," suggesting potential instability or areas needing refinement. While it supports multiple languages, the primary focus and default models appear to be Chinese.