russian_asr

This is a test line for checking github features.

This is a university project for automatic speech recognition in russian, using the Nvidia NeMo toolkit.

Dataset(s) used for training:

• Mozzila Common Voice, Russian Language (https://drive.google.com/drive/folders/1HJbaEVoK_i__8vqOVwcYyYIiWEfwQUsD?usp=sharing)

Best WER achieved on AN4 dataset using base config modified by:

• lr: 0.02
• weight decay: 0.005
• epoch: 200
• WER: 0.1255 = 12.55 %