Indian language identification using time-frequency texture features and kernel ELM

Precise identification of the language from speech utterance is a prime task of a language identification system and has been extensively utilized in multilanguage speech applications. This article presents Indian language identification system using textural descriptors extracted from time-frequenc...

Full description

Saved in:
Bibliographic Details
Published in:Journal of ambient intelligence and humanized computing 2023-10, Vol.14 (10), p.13237-13250
Main Authors: Birajdar, Gajanan K., Raveendran, Smitha
Format: Article
Language:English
Subjects:
Accuracy
Algorithms
Artificial Intelligence
Artificial neural networks
Classification
Computational Intelligence
Engineering
Feature extraction
Identification systems
Knowledge management
Language
Linear phase
Machine learning
Neural networks
Original Research
Representations
Robotics and Automation
Speech
Texture
Time-frequency analysis
User Interfaces and Human Computer Interaction
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Precise identification of the language from speech utterance is a prime task of a language identification system and has been extensively utilized in multilanguage speech applications. This article presents Indian language identification system using textural descriptors extracted from time-frequency visual representation. The conventional LPC and MFCC feature extraction approaches for language identification have limited detection accuracy. In the first step, an input speech signal is converted into spectrogram, MFCC and cochleagram images representation. These speech sample visual representations can be treated as a texture image characterizing energy variations in different frequency-bands over time. Second step comprises extraction of completed linear binary pattern (CLBP), linear phase quantization (LPQ) and Weber local descriptor (WLD) textural features from visual representations. Finally, the kernel extreme learning machine (KELM) classifier has been employed for the language specific class label identification. The proposed algorithm validation is carried out using the IIIT-H Indic speech databases incorporating seven Indian languages from Indo-Aryan and Dravidian family. It is evident from the experimental results that the proposed time-frequency texture descriptor method outperforms other machine learning algorithms.
ISSN:1868-5137
1868-5145
DOI:10.1007/s12652-022-03781-5