Scholarlink

In the field of computer science speech recognition play vital role. Speech is nothing but primary means of communication in the field of speech recognition. This paper described a Framework for identifying speakers of isolated vocabulary of Ahirani spoken language. In this research article main focused is on Ahirani spoken language which is spoken in Maharashtra state from Khandesh Region. Speech database is collected from native speakers. Database is consisting of 300 Spoken Ahirani isolated words. Hidden Markov Model is used with Gaussian Mixture Model. Mel – Frequency Cepstral Coefficient (MFCC) is used as speech feature extraction method. Total 39 MFCC features are extracted. Using HMM mixing with Gaussian mixture is the first step towards to develop IVRS system for Spoken language. Experimental work is performed using four native speakers. Model is trained in speaker dependent environments. Word error rate is calculated 3.46%.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Automatic Speech Recognition System for Ahirani Language Using HMM in Speaker Dependent Environment

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Sunil Patil,
  • Chandrashekhar H. Patil,
  • Meenal Jabde,
  • Vikas Mahandule

摘要

In the field of computer science speech recognition play vital role. Speech is nothing but primary means of communication in the field of speech recognition. This paper described a Framework for identifying speakers of isolated vocabulary of Ahirani spoken language. In this research article main focused is on Ahirani spoken language which is spoken in Maharashtra state from Khandesh Region. Speech database is collected from native speakers. Database is consisting of 300 Spoken Ahirani isolated words. Hidden Markov Model is used with Gaussian Mixture Model. Mel – Frequency Cepstral Coefficient (MFCC) is used as speech feature extraction method. Total 39 MFCC features are extracted. Using HMM mixing with Gaussian mixture is the first step towards to develop IVRS system for Spoken language. Experimental work is performed using four native speakers. Model is trained in speaker dependent environments. Word error rate is calculated 3.46%.