Multiband Spectral Subtraction dengan Menggunakan Gaussian Window untuk Meningkatkan Kualitas Sinyal Ucapan Berderau
Abstract
The presence of noise damages the quality and intelligibility of speech signals, thereby reduces the performance of speech based applications. Noise is unavoidable and unpredictable. The most common type of noise found in the environment is non stationary noise. Non-stationary noise does not affect the speech signal uniformly over whole frequencies. Therefore conventional spectral subtraction methods are not effective. This paper, propose a Multiband spectral subtraction method which is divides the noisy speech signal into several frequency bands uniformly and reduce spectral noise in each band independently. The number of frequency bands tested are 4, 8 and 12 bands. The window type affects spectral estimates. Therefore this method is tested for three types of windows, hamming, sine multitapper with number of tapper 4 and Gaussian. The performance of the proposed method is assessed based on the PESQ of the enhanced speech signal. The Gaussian window with side lobe attenuation 3.75 had the best PESQ in all number of frequency bands compared to hamming and sine multitaper. The best average PESQ is obtained using Multiband Spectral subtraction using a Gaussian window with side lobe attenuation 3.75 and number of bands 4.
Keywords: Multiband Spectral Subtraction, Gaussian Window dan PESQ
Abstrak
Keberadaan derau merusak kualitas dan kejelasan sinyal ucapan sehingga mengurangi kinerja aplikasi yang berbasis ucapan. Derau tidak dapat dihindari dan tidak terprediksi. Jenis derau yang paling umum ditemukan pada lingkungan adalah derau non stasioner. Derau non stasioner memberi pengaruh secara tidak seragam pada semua frekuensi sinyal ucapan. Oleh karena itu metoda pengurangan spektral yang konvensional menjadi tidak efektif. Pada penelitian ini diusulkan Multiband spectral subtraction. Metoda ini membagi sinyal menjadi beberapa band frekuensi secara seragam dan melakukan pengurangan spektral derau secara saling bebas pada setiap band. Jumlah band frekuensi yang diujikan yaitu 4, 8 dan 12 band. Jenis window mempengaruhi estimasi spektral. Oleh karena itu ini diujikan pada tiga jenis window yaitu hamming, sine multitapper dengan jumlah tapper 4, dan gaussian window. Kinerja yang diusulkan dinilai berdasarkan nilai PESQ dari sinyal ucapan hasil perbaikan. Berdasarkan hasil pengujian didapatkan bahwa penggunaan gaussian window dengan redaman side lobe 3,75 memiliki nilai PESQ terbaik disemua jumlah band frekuensi dibandingkan hamming dan sine multitapper. Nilai rata-rata PESQ terbaik diperoleh dengan menggunakan Multiband Spectral subtraction menggunakan gaussian window dengan redaman side lobe 3,75 dan jumlah band 4.
Kata kunci: Multiband Spectral Subtraction, Gaussian Window and PESQ
Keywords
Full Text:
PDF (Indonesian)References
Pardede., Hilman, F. Nonlinear Spectral Subtraction Based on Tsallis Statistics for Speech Enhancement. 2013. INKOM Vol. 7 No 1 Mei: artikel 228.
S. F. Boll. Suppression of Acoustic Noise in Speech Using Spectral Subtraction. 1979. IEEE Trans. Acoust: Speech Signal Process Vol. 27 No. 2: 113– 120.
Kamath, S., Loizou, P. A Multiband Spectral Subtraction Methods for Enhancing Speech Corrupted by Colored Noise. 2002. Proc. IEEE Intl. Conf. Acoustics, Speech, Signal Processing.
M. Berouti., R. Schwartz., J. Makhoul. Enhancement of Speech Corrupted by Acoustic Noise. 1979. Proc. IEEE ICASSP: 208-211.
Supriya, P. Sarvade., Dr. Shridhar, K., Varun, P. Sarvade. Multi-Band Spectral Subtraction for Speech Enhancement Using Sine Multitaper. 2016. IOSR Journal of VLSI and Signal Processing (IOSR-JVSP) Vol 6 Issue 6 Ver. II Nov – Dec.
S. China, Venkateswarlu., A. Subba Rami Reddy, K. Satya Prasad. Speech Enhancement Using Boll’s Spectral Subtraction Method Based on Gaussian Window. 2014. Global Journal of Researches in Engineering: F Electrical and Electronics Engineering Vol 14 Issue 6 Version 1.0.
Navneet, Upadhyay., Abhijit, Karmakar. Single-Channel Speech Enhancement using Critical-Band Rate Scale Based Improved Multi-Band Spectral Subtraction. 2013. Journal of Signal and Information Processing ISSN 2159-4465 Vol 04: 314.
Venkata Rami Reddy Datla. Implementation and Evaluation of Spectral Subtraction (SS) with Minimum Statistics and Wiener Beamformer Combination. 2013. Master Thesis of Electrical Engineering, School of Electrical Engineering, Blekinge Institute of Technology (BTH), Sweden.
Scoot, Pennock. Accuracy of the Perceptual Evaluation of Speech Quality Algorithm. 2017. Lucent Technologies.
DOI: http://dx.doi.org/10.36055/setrum.v8i1.5469
Refbacks
- There are currently no refbacks.