hama on-device NLP

Acknowledgements

hama exists thanks to collaborative efforts across research, engineering, and community sponsors. We are grateful for the following contributions.

Core maintainers

Thank you to the maintainers who keep hama open source.

Community

Thank you to everyone who reports issues, shares ideas, and helps steer the project.

Training data

G2P

  • English: CMUDict, IPADict, and English Wikipron.
  • Korean: Korean Wikipron.

  • Chinese (Hans/Hant), Dutch, French, German, Italian, Japanese, Portuguese, Spanish, and Thai: Wikipron TSV exports used in the multilingual manifests.

ASR

  • LibriSpeech, including the train-clean-100, train-clean-360, and train-other-500 training splits.
  • Mozilla Common Voice English.
  • AMI meeting speech via the open ami split from esb/datasets.

Sponsors

  • Sunah Hong — donated an NVIDIA DGX Spark for training and benchmarking.

Upstream projects

  • ONNX Runtime for powering cross-platform inference.
  • Bun, Node, and the Python ecosystem for reliable tooling.
  • Astro for the content/workflow experience.
  • 나눔 손글씨 김유이체 for the logo wordmark font, courtesy of NAVER CLOVA.

If you or your organization rely on hama, support the project via GitHub Sponsors. For business inquiries, reach out at intelligent.hama@gmail.com.