A real time speech to text conversion technique for Bengali language

Abstract

This paper presents a model to convert natural Bengali language to text. The proposed model requires the usage of the open sourced framework Sphinx 4 which is written in Java and provides the required procedural coding tools to develop an acoustic model for a custom language like Bengali. Our main objective was to ensure that the system was adequately trained on a word by word basis from various speakers so that it could recognize new speakers fluently. We used a free digital audio workstation (DAW) called Audacity to manipulate the collected recording data via continuous frequency profiling techniques to reduce the Signal-to-Noise-Ratio (SNR), vocal leveling, normalization and syllable splitting as well as merging which ensure an error free 1:1-word mapping of each utterance with its mirror transcription file text. To evaluate the performance of proposed model, we utilize an audio dataset of recorded speech data from 10 individual speakers consisting of both males and females using custom transcript files that we wrote. Experimental results demonstrate that the proposed model exhibits average 71.7% accuracy for our tested dataset.