Baselines are built using Kaldi. Please see this README file (opens in new tab) for instructions on how to replicate the baselines and for links to lexicons for all languages.
The Word Error Rates of the baseline systems for all three languages are below:
Language | GMM-HMM | DNN | TDNN |
Tamil | 33.55 | 25.47 | 19.45 |
Telugu | 40.12 | 34.97 | 22.61 |
Gujarati | 23.78 | 27.79 | 19.76 |