This repository is a Python implementation for HMM-DNN model which is a deep learning model in speech recognition. First, we use HMM-GMM model for labeling an existing speech data. Then, we would use this labeled data for training the HMM-DNN model. Also, we use MLP as for the DNN part of the model.

Clone the program.

git clone https://github.com/raminnakhli/HMM-DNN-Speech-Recognition.git

The requirements are some common packages in machine learning. You can install them using below command.

pip install -r requirement.txt

We use an Arabic speech dataset which is available on https://archive.ics.uci.edu/ml/datasets/Spoken+Arabic+Digit. For cloning the dataset, you can use the download_data.sh file. This file creates a data directory and downloads the dataset in it.

Now, you can run the model with default configuration using the below command.

python main.py

Ramin Ebrahim Nakhli - raminnakhli@gmail.com

Project Link: https://github.com/raminnakhli/HMM-DNN-Speech-Recognition