We found a match
Your institution may have access to this item. Find your institution then sign in to continue.
- Title
Enhanced Dialectal Speech Recognition in Punjabi Using Pitch-based Acoustic Modeling.
- Authors
Bhardwaj, Vivek; Thakur, Deepak; Gera, Tanya; Sharma, Vikrant
- Abstract
Automatic Speech Recognition (ASR) systems usually have difficulty accurately transcribing dialectal variations, resulting in subpar performance in areas where dialectal variants are common. The pitch-based Dialect ASR method we described in this paper aims to improve voice recognition for dialectal differences of Punjabi language. We use the pitch information that was taken out of the voice signal as a feature to enhance the dialectal nuance recognition. The suggested system includes a cutting-edge pitch-based feature extraction module that records minute differences in pitch patterns linked to various dialects. This module gives the ASR system the ability to distinguish between phonetic units more effectively and faithfully depict the distinguishing features of dialectal speech. To develop reliable representations from the pitch-based data, we also use deep learning approaches, speaker adaptive training, and vocal-tract length normalization (VTLN). The experimental results show the significant reduction in the WER of 6.63% and 4.98% for Malwa and Majha dialects. Language learning applications could benefit from the developed Punjabi dialectal speech recognition system by offering learners exposure to various dialects and accents. This can help learners develop a well-rounded understanding of the language and better adapt to different regional variations.
- Subjects
DEEP learning; AUTOMATIC speech recognition; SPEECH perception; ACOUSTIC models; SPEECH; FEATURE extraction
- Publication
Ingénierie des Systèmes d'Information, 2023, Vol 28, Issue 6, p1557
- ISSN
1633-1311
- Publication type
Article
- DOI
10.18280/isi.280612