L-A-Sandhu/Speech-commands_keyword-spotting

Speech Command Recognition

A Keras implementation of neural attention model for speech command recognition

This repository presents a recurrent attention model designed to identify keywords in short segments of audio. It has been tested using the Google Speech Command Datasets (v1 and v2). For a complete description of the architecture, please refer to the (https://arxiv.org/abs/1808.08929). The Demo notebook is preconfigured with a set of tasks: ['12cmd', 'leftright', '35word', '20cmd']. Each of these refer to how many commands should be recognized by the model. When loading the Google Speech Dataset, the user should also select which version to download and use by adjusting the following line:

gscInfo, nCategs = SpeechDownloader.PrepareGoogleSpeechCmd(version=1, task = '35word')

The rest of the repository is divided as follows.

• Requirements
• Cloning this repository
• Pre-trained Model
• Coustom Traning

Requirements

Install the requirements using

pip install -r requirements.txt

Cloning this repository

• Download or clone this repository;
• Open the Demo notebook;
• Choose how many words should be recognized and the Google Speech Dataset version to use;
• Run training and tests.

Pre-trained Model

If you want a pretrained model, model-attRNN.h5 contains pre-trained weights for task 35word, version=2.

Coustom Traning

If you want to train with your own data:

• Use the audioUtily.py WAV2Numpy function to save your WAV files in numpy format. This speeds up loading considerably;
• Create a list_IDs array containing the paths to all the numpy files and a labels array with corresponding labels (already converted to integers);
• Instantiate a SpeechGenerator.py SpeechGen class;
• Create your own Keras model for audio classification or use one provided in SpeechModels.py;
• Train the model.