A Novel Approach for Automatic Speaker Identification and Word Identification of Tai-Phake Speakers from short-duration spoken words using MMCCT and Feed-Forward Neural Network

Abstract

Objective: Speaker identification and word identification are critical tasks in language processing and artificial intelligence. This research focuses on applying feed-forward neural networks to the specific context of the Tai-Phake language, an endangered language spoken in parts of Northeast India. The primary objective is to develop effective models for both speaker identification and word identification in Tai-Phake, leveraging the capabilities of neural networks. Methods: The study begins by collecting a corpus of Tai-Phake speech data from 18 speakers recording 50 different words 10 times each, which is annotated and pre-processed to facilitate model training. For speaker recognition, a feed-forward neural network architecture is designed to accurately identify and authenticate individuals based on their unique vocal characteristics. The model's performance is evaluated using metrics such as accuracy, precision, recall, and F1 score. Additionally, for word recognition, another neural network model is developed to accurately identify spoken Tai-Phake utterances. This involves training the model on a labeled dataset of spoken words, optimizing it for robust performance across various dialectal variations and speaking styles within the Tai-Phake community. Findings: Experimental results demonstrate the effectiveness of the proposed neural network models in achieving high accuracy rates for both speaker and word recognition tasks in Tai-Phake. The implications of this research extend to the preservation and documentation of endangered languages, showcasing how advanced machine learning techniques can contribute to linguistic research and cultural heritage preservation efforts. Novelty: No prior work has been done in Tai-phake speaker identification and word identification using a combination of Feature Fusion and Neural Networks.