Please use this identifier to cite or link to this item:
http://hdl.handle.net/10603/265831
Title: | Automatic Speech Recognition of Indian Languages using Soft Computing |
Researcher: | Agnihotri Prashant Prakashrao |
Guide(s): | Khanale P. B. |
Keywords: | Computer science |
University: | Swami Ramanand Teerth Marathwada University |
Completed Date: | 30/01/2019 |
Abstract: | Speech is the most common and effective medium of communication for information newlineexchange between human beings. People often need to use machines for various purposes newlinelike controlling applications through commands, data entry and its storage, document newlinepreparation, data analysis, information retrieval, entertainment etc. Automatic recognition newlineof speech is a process of conversion of acoustic signal of speech utterances into the text. newlineIn this study, the process of automatic speech recognition of Indian languages newlineincludes three phases: First phase: Development of noise free speech database of Indian newlinelanguages like Marathi and Hindi, Second phase: Ensemble Feature extraction and Third newlinephase: classification. newlineIn first phase, three speakers were selected to record the speech corpora of Marathi newlineand Hindi language having different age group, different socio-linguistic background and newlinedifferent gender. The laptop and uni-condenser microphone was used to record the speech newlinecorpora for both languages which may cause addition of noise into the speech corpora. newlineThe filtering techniques like pre-emphasis filter, butterworth filter like low pass, newlinehigh pass, band pass and band stop were applied on the speech databases of Marathi and newlineHindi languages to remove the noise from recorded signal. Once speech signals have been newlinefiltered through filtering techniques, it is essential to know which method is more newlineefficient than others to reduce the noise level from the speech signal. Hence the signal to newlinenoise ratio (SNR) and Information to Entropy Ratio (IER) have been determined for each newlinefiltering techniques. As SNR/IER is inversely proportional to noise power of the newlinesignal/Entropy of signal, it is observed that band stop filtering technique has maximum newlinevalue for SNR/IER for the vowels of Marathi and Hindi languages which indicate that newlineless noise is present in the signal. Band stop filtering technique is also applied on newlineconsonants of Marathi and Hindi languages. The filtered signals are stored into wave files newlineto form noise free speech databases |
Pagination: | 129p |
URI: | http://hdl.handle.net/10603/265831 |
Appears in Departments: | School of Computational Sciences |
Files in This Item:
File | Description | Size | Format | |
---|---|---|---|---|
01_title.pdf | Attached File | 107.66 kB | Adobe PDF | View/Open |
02_certificate.pdf | 350.32 kB | Adobe PDF | View/Open | |
03_abstract.pdf | 354.11 kB | Adobe PDF | View/Open | |
04_declaration.pdf | 346.78 kB | Adobe PDF | View/Open | |
05_acknowledgement.pdf | 354.65 kB | Adobe PDF | View/Open | |
06_content.pdf | 475.84 kB | Adobe PDF | View/Open | |
07_list_of_tables.pdf | 353.51 kB | Adobe PDF | View/Open | |
08_list_of_figures.pdf | 363.88 kB | Adobe PDF | View/Open | |
09_abrivations.pdf | 348.57 kB | Adobe PDF | View/Open | |
10_chapter 1.pdf | 3.18 MB | Adobe PDF | View/Open | |
11_chapter 2.pdf | 6.72 MB | Adobe PDF | View/Open | |
12_chapter 3.pdf | 5.69 MB | Adobe PDF | View/Open | |
13_chapter 4.pdf | 6.28 MB | Adobe PDF | View/Open | |
14_conclusiion.pdf | 2.76 MB | Adobe PDF | View/Open | |
15_summary.pdf | 473.55 kB | Adobe PDF | View/Open |
Items in Shodhganga are licensed under Creative Commons Licence Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0).
Altmetric Badge: