nrishabh/adapted-hi-asr-model

Adaptation of Hindi Speech Recognition Model (using Kaldi)

Creating a Hindi Speech Recognition Model using Kaldi

Kaldi Toolkit

Before running anything, make sure you download and install kaldi-toolkit.

Hindi Datasets

The following datasets have been found for Hindi language.

Common Voice

OpenSLR: Multilingual and code-switching ASR Challenge Dataset (Sub-task 1)

OpenSLR: Multilingual and code-switching ASR Challenge Dataset (Sub-task 2)

Speech Lab, IITM - Hindi Corpus

Structuring Dataset into IITM-ASR-Toolkit Folder

Replicate the following directory structure to develop the model.

dataset
├── ...
├── audio
|    ├── train: Contains all WAV files for training data
|    ├── dev: Contains all WAV files for development data
|    └── test: Contains all WAV files for testing data
├── data
|    ├── train: Contains all transcription files for training data
|    ├── dev: Contains all transcription files for development data
|    ├── test: Contains all transcription files for testing data
|    └── local
|       └── dictionary: Contains all phonetic dictionary files for the model
└── ...