FormoSpeech (Speech in Formosa)

Advancing Speech AI for the Linguistic Diversity of Taiwan

About Us

Welcome to FormoSpeech (Speech in Formosa), an open-source initiative and community hub hosted on Hugging Face. We are dedicated to the development and promotion of speech and language technologies for the languages of Taiwan.

This organization is founded by Hung-Shin Lee (李鴻欣) at NTNU, with support from Li-Wei Chen (陳力瑋) at Voxmosa. It aims to bring together researchers, developers, and language enthusiasts to build and share high-quality, accessible resources for the Taiwanese linguistic context.

Our Mission

Our primary goal is to create and curate foundational models, datasets, and tools to support the rich linguistic tapestry of Taiwan. We believe that advancing AI capabilities for these languages is crucial for digital inclusion, cultural preservation, and future innovation.

We are specifically focused on:

• Taiwanese Mandarin (國語): Developing models that capture the unique accent, lexicon, and nuances of Mandarin spoken in Taiwan.
• Taiwan Taigi (臺灣台語): Building resources for a vital language with a rich oral tradition but historically fewer digital resources.
• Taiwan Hakka (臺灣客語): Supporting the various dialects of Hakka spoken across Taiwan.
• Formosan Languages (臺灣原住民語): Contributing to the digital preservation and revitalization of Taiwan's Formosan languages.

License and Data Access

To ensure the ethical use and sustainable development of our resources, please adhere to the following guidelines:

Models: All models released by FormoSpeech are licensed under Creative Commons Attribution-NonCommercial (CC BY-NC). They are strictly for non-commercial use.

Datasets: Our datasets are currently used ONLY for internal academic research and remain restricted to ensure privacy and proper usage. Please note that although access requests are currently enabled for most datasets, they will NOT be granted or released at this time. If you have special inquiries, exceptional needs, or feedback, please feel free to reach out to us via email.

What You'll Find Here

This organization will host a growing collection of:

• Models: Pre-trained models for tasks such as Automatic Speech Recognition (ASR), Text-to-Speech (TTS), Machine Translation (MT), speaker/language identification, and more, all fine-tuned or trained on Taiwanese language data.
• Datasets: Curated and pre-processed speech corpora suitable for training and evaluating speech AI models.
• Demos & Tools: Interactive demos (Spaces) and tools to facilitate research and application development.

Citation

If you use any models, datasets, or demos from FormoSpeech in your research or projects, we would be incredibly grateful if you could cite our organization.

You can use the following BibTeX entry:

@misc{formospeech,
  author       = {Hung-Shin Lee and Li-Wei Chen},
  title        = {FormoSpeech: Advancing Speech AI for the Linguistic Diversity of Taiwan},
  year         = {2025},
  publisher    = {Hugging Face},
  howpublished = {\url{[https://huggingface.co/FormoSpeech](https://huggingface.co/FormoSpeech)}}
}