Simple pytorch implementation of jamo-based korean speech recognition.
We used KsponSpeech corpus dataset for training/validation.
We referred to [2]'s preprocessing method and [3]'s model structure. Thank you!
- Write Description
- Change evaluation metrics (L0 norm --> Levenshtein distance for jamo)
- Fine tuning & code cleanup
[1] Dataset
[2] https://github.com/sooftware/End-to-end-Speech-Recognition
[3] https://github.com/clovaai/ClovaCall
[4] https://github.com/kaniblu/hangul-utils
[5] Chan, William, et al. "Listen, attend and spell." arXiv preprint arXiv:1508.01211 (2015)