語音辨識被廣泛被運用在日常生活中，智慧型手機的辨識系統就是一個相當好的例子。然而語音辨識系統已經發展多年，早在1970年代開始便有許多學者針對此議題發展出各種不同的方法及不錯的成果，近年來針對日常生活中的噪音干擾更是有學者提出時頻感知域來應對，發展出一套辨識參數。
本論文即是採用時頻感知域參數，在本文中提出2套語音訊號的處理方法，一個是針對噪音環境下的語音增強，另一個則是在噪音環境下的語音辨識。增強部分除了使用時頻域多從解析外更搭配溫妮濾波器，舉例而言，在訊雜比為-10 dB時我們提出的增強方法可以將訊號增強到0.28 dB且在訊雜比為0 dB時產出的結果可以增進到7.11dB跟其他相關研究比較可說是較好的進步。另一個辨識部分，採用時頻分析域和傳統的梅爾濾波器做結合，在乾淨語音中辨識率可以從只有梅爾係數的68.62%，增進到83.10%，且我們所提出的參數針對特定容易混淆的字元，提供了另一種另外判別的方法，使彼此之間較不容易辨識錯誤。此外我們考量現實環境中，噪音所帶來的影響，而我們所提出的參數皆有對抗不同噪音的穩定性。

Speech recognition has been widely used in daily life, such as the recognition system in smart phones. Since 1970s a lot of recognition methods have been proposed and many of them also achieved high recognition rate. However, by considering the practical situation, noisy environments might need to be taken in to account. In recent years researches on Spectral-Temporal Receptive Fields (STRF) developed for recognition.
In this paper, STRF was further studied and applied to two applications. One was for speech enhancement and, the other was for speech recognition. For speech enhancement, the proposed method utilized STRF analysis and wiener filtering to improve the speech quality. In the noisy environment with the white noise -10 and 0 dB level, our proposed method achieved 0.28 dB and 7.11 dB respectively. Compared with other studies it is a significant improvement.
As for the speech recognition, the proposed method combined the conventional parameters MFCC and STRF features including rate and scale. In clean speech the MFCC recognition rate was about 68.62% and our proposed features could obtain 83.10% recognition rate. In addition, by considering the real-world environment that included, the impact of noise, the experimental results also showed higher recognition rates at each noise level.

參考文獻
[1] T.S. Chi, P. Ru and S. Shamma, “Multiresolution spectrotemporal analysis of complex sounds,” , Journal of the Acoustical Society of America, vol. 118, no. 2, pp 887-906, 2005.
[2] Available on: http://neural.cs.washington.edu/
[3] Z.M. Dan and F.S. Monica, “A study about MFCC relevance in emotion classification for SRoL database,” Electrical and Electronics Engineering (ISEEE ,pp.1,4, 11-13 Oct. 2013.
[4] M. Lagunas and A.I. Perez-Neira,”Spectral subtraction and spectral estimation,” IEEE Workshop on Statistical Signal and Array Processing Proceedings of the Tenth, pp.383-386, 2000.
[5] 結合雜訊抑制與帶聲語音重建之語音增強系統，廖育志。
[6] PESQ Introduction http://www.pal-acoustics.com/index.php?a=services&id=143&lang=cn
[7] L. Rabiner, “A tutorial on hidden Markov models and selected applications in speech recognition,” Proceedings of the IEEE , vol.77, no.2, pp.257,286, Feb 1989.
[8] 語音訊號與系統，王小川，第二版，全華圖書，民國98年
[9] C.C. Hsu, T.E Lin, J.H. Chen and T.S. Chi, “Spectro-temporal subband Wiener filter for speech enhancement,” IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp.4001,4004, 25-30 March 2012
[10] J. Woojay and B. H. Juang, “Speech analysis in a model of the central auditory system,” IEEE Transactions on Audio, Speech, and Language Processing, vol. 15, no. 6, pp.1802-1817, Aug. 2008.
[11] X. Yang, K. Wang, and S. A. Shamma, “Auditory representations of acoustic signals,” IEEE Trans. Inf. Theory, vol. 38, no. 2, pp. 824–839, Mar. 1992.
[12] Y. Ephraim, “Gain-adapted hidden markov models for recognition of clean and noisy speech,” IEEE Trans. Signal Process., vol. 40, no. 6,pp. 1303–1316, Jun. 1992.
[13] C. Corredor-Ardoy, L. Lamel, M. Adda-Decker and J-L Gauvain, “Multilingual phone recognition of spontaneous telephone speech,” IEEE International Conference on Acoustics, Speech and Signal Processing, vol.1, no., pp.413,416 vol.1, 12-15 May 1998
[14] Chunghwa Digits Database. [Online] Available: http://www.aclclp.org.tw/use_mat_c.php
[15] X. Q. Zhao and J. Wang, “A new noisy speech recognition method,” in Proc. IEEE International Symposium on Communications and Information Technology, Oct. 2005, pp.292-296.
[16] S. H. Choi, H. K. Kim, and H. S. Lee, “Speech recognition using quantized LSP parameters and their transformations in digital communication,” Speech Communication, vol. 30, pp. 223-233, Apr. 2000.
[17] R. Lyon and S. Shamma, “Auditory representation of timbre and pitch,” Journal of Auditory Computations, pp 221–270.
[18] W. C. Lin, H. T. Fan and J. W. Hung, “DCT-based processing of dynamic features for robust speech recognition,” in Proc. IEEE International Symposium on Chinese Spoken Language Processing, Nov. 2010, pp.12-17.
[19] V. F. S. Alencar and A. Alcaim, “LSF and LPC - derived features for large vocabulary distributed continuous speech recognition in Brazilian Portuguese,” in Proc. Asilomar Conference on Signals, Systems and Computers, Oct. 2008, pp. 1237-1241.
[20] M. A. Ferrer, I. G. Alonso, and C. M. Travieso, “Influence of initialisation and stop criteria on HMM based recognisers,” Electronics Letters, vol.36, no.13, pp.1165-1166, Jun. 2000.
[21] D. Sarkar, “Randomness in generalization ability: a source to improve it,” IEEE Transactions on Neural Networks, vol.7, no.3, pp.676-685, May 1996.