Introduction

This project focuses on automatic speech recognition task, specifically speech transcription, using Deep Neural Network (DNN) architecture. The model was trained and test on 10% of train-clean-100 and test-clean from Librispeech. The implementation refered to AssemblyAI tutorial on E2E speech recognition system

Model architecture

This project employs CRNN structure with convolutional and GRU blocks to process the input spectrogram. The model output the prediction probabilities of the letters over the time steps.

Installation

To run the code, you need python, pytorch, and numpy

How to run

asr_main.py incorperates the training loop and the testing stage of the speech transcription model

Authors

• Diep Luong
• Fareeda Mohammad