Berauti Spectral Subtraction dengan Gaussian Window untuk Peningkatan Akurasi Pengenalan Ucapan Berderau
Main Article Content
Keywords
Abstract
The accuracy of speech recognition system decreases when used on a noisy speech. Therefore, the speech recognition system needs to be supported by a speech enhancement method. This study proposes Berauti spectral subtraction method that uses gaussian window and minimum statistics noise estimation in order to improve the quality of noisy speech hence increase the accuracy of noisy speech recognition. Speech recognition system is built using the Hidden Markov Model Toolkit (HTK). This study applied three types of noise, five SNR levels, six oversubtraction values and four sidelobe gaussian window attenuation values with 1500 speech signals. Improvement of speech recognition accuracy using Gaussian window is compared with Hamming window. The results of the study shows that sidelobe and oversubtraction attenuation values affects recognition accuracy. The average speech recognition accuracy using gaussian window improve about 36.4% which is obtained at oversubtraction 4.75 and sidelobe attenuation = 1.5. Whereas, application of hamming window improves the accuracy about 18,7 % which is obtained at oversubtraction 2.5. Spectral subtraction using gaussian window or hamming window is able to improve the speech recognition accuracy, but gaussian window is better than hamming window.
Keywords : Berauti spectral subtraction, gaussian window, speech recognition
Abstrak
Akurasi sistem pengenalan ucapan menurun ketika digunakan pada ucapan berderau. Oleh karena itu, sistem pengenalan ucapan perlu didukung dengan metoda perbaikan sinyal ucapan. Pada penelitian ini diusulkan metoda Berauti spectral subtraction yang menerapkan gaussian window dan estimasi derau minimum statistik untuk memperbaiki kualitas sinyal berderau sehingga dapat meningkatkan akurasi pengenalan ucapan berderau. Sistem pengenalan ucapan dibangun menggunakan Hidden Markov Model ToolKit (HTK). Pada penelitian ini divariasikan tiga jenis derau, lima level SNR, enam nilai oversubtraction dan empat nilai redaman sidelobe gaussian window dengan 1500 sinyal ucapan. Peningkatan akurasi pengenalan ucapan yang menggunakan gaussian window dibandingkan dengan hamming window. Hasil penelitian ini menunjukan pemilihan nilai redaman sidelobe dan oversubtraction mempengaruhi akurasi pengenalan. Rata-rata peningkatan akurasi pengenalan ucapan sebesar 36,4 % diperoleh pada nilai oversubtraction 4.75 dan redaman sidelobe 1.5. Penggunaan hamming window memiliki rata-rata peningkatan akurasi pengenalan sebesar 18,7 % pada nilai oversubtraction 2.5. Metoda spectral subtraction yang menggunakan gaussian window atau hamming window, keduanya mampu menaikan akurasi pengenalan ucapan, akan tetapi gaussian window memiliki hasil yang lebih baik dibanding hamming window
Kata Kunci : Berauti spectral subtraction, gaussian window, pengenalan ucapan
References
Xiao Xiong. Speech Enhancement with Applications in Speech Recognition. Nanyang Technological University, 2006
Gales, Mark John Francis. "Model-based techniques for noise robust speech recognition." PhD diss., University of Cambridge, 1995.
Arun Narayanan, Xiaojia Zhao., DeLiang Wang, Eric Fosler-Lussier. "Robust Speech Recognitiion Using Multiple Prior Models For Reconstructon". ICASSP - IEEE. 2011;4800-48003
Boll: ”Suppression of acoustic noise in speech using spectral subtraction”, IEEE Trans. Acoust. Speech and Sig. Proc., 27:113-120, 1979.
M. Berouti, R. Schwartz, J. Makhoul, ”Enhancement of speech corrupted by acoustic noise”, Proc. IEEE ICASSP, 208-211, 1979.
Harald Gustafsson, Sven Nordholm and Ingvar Claesson, "Spectral Subtraction With Adaptive Averaging Of The Gain Function", 6th European Conference on Speech Communication and Technology (EUROSPEECH’99) Budapest, Hungary, September 5-9, 1999
S. China Venkateswarlu, A. Subba Rami Reddy & K. Satya Prasad, "Speech Enhancement using Boll’s Spectral Subtraction Method based on Gaussian Window", Global Journal of Researches in Engineering: Electrical and Electronics Engineering Vol. 14 Issue 6 Version 1.0 2014
A. M. Kandoz, Digital speech, 2nd edition, Willey, 2002.
Martin, Rainer. "Noise power spectral density estimation based on optimal smoothing and minimum statistics." IEEE Transactions on speech and audio processing Vol.9 No.5 pp.504-512, 2001.
Urmila Shrawankar, Vilas Thakare. An "Adaptive Methodology for Ubiquitous ASR System", Computer and Information Science, vol 6 no.1, pp 58-69, 2013
Fitrilina ., Rahmadi Kurnia, Siska Aulia, "Pengenalan Ucapan Metoda MFCC-HMM Untuk Perintah Gerak Robot Mobil Penjejak Identifikasi Warna", Jurnal Nasional Teknik Elektro (JNTE) Vol 2,No 1: Maret 2013
Fitrilina, "Sistem Pengenalan Isolated Digit yang Robust Dengan Menggunakan Spectral Subtraction Berdasarkan Minimum Statistics", Jurnal Teknika No 35 Vol 1. 2011,
Venkata Rami Reddy Datla, "Implementation and evaluation of spectral subtraction (SS) with minimum statistics and wiener beamformer combination". Master Thesis of Electrical Engineering, school of Engineeringctrical Engineering ,Blekinge Institute of Technology (BTH), Sweden. 2013
Martin, Rainer. "Spectral subtraction based on minimum statistics." Power V6 n.8, 1994
Bhatnagar, Mukul, “Modified Spectral Subtraction Method Combine With Percetual Weighting for Speech Enhancement”, Thesis, The University of Texas, Dallas, 2002
Hilman F. Pardede, "Nonlinear Spectral Subtraction Berbasis Tsallis Statistics untuk Peningkatan Kualitas Sinyal Ucapan", INKOM, Vol. 7, No. 1, 2013.
Article Sidebar
Article Details
Authors who publish with this journal agree to the following terms:
- Authors retain copyright and grant the journal right of first publication with the work simultaneously licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License that allows others to share the work with an acknowledgement of the work's authorship and initial publication in this journal.
- Authors are able to enter into separate, additional contractual arrangements for the non-exclusive distribution of the journal's published version of the work (e.g., post it to an institutional repository or publish it in a book), with an acknowledgement of its initial publication in this journal.
- Authors are permitted and encouraged to post their work online (e.g., in institutional repositories or on their website) prior to and during the submission process, as it can lead to productive exchanges, as well as earlier and greater citation of published work (See The Effect of Open Access).
The authors declare that:
1. This paper has not been published in the same form elsewhere.
2. It will not be submitted anywhere else for publication prior to acceptance/rejection by this Journal.
3. A copyright permission is obtained for materials published elsewhere and which require this permission for reproduction.