Your browser does not support JavaScript!
智慧型信號處理實驗室(Intelligent Signal Processing)
不怕沒得學,就怕你不肯學,勇敢築夢!!
繁體

Recent

數據載入中...
讓機器人跟你交談- 淺談訊源定位及隱蔽訊號分離技術

全自動的機器人助理是許多電影常見的場景,它可以解讀並執行我們的口語命令也可以跟我們對談。現實生活中,工程師們也正努力將它實現。如何讓機器人正確地接收你的口語信號是首要問題,其實要讓機器人像我們一樣可以精確地判斷聲音來源,不是一件容易的事,工程上稱為訊源定位(Source Localization)。實際生活上,常見的應用為雷達掃描定位、地震震源定位、衛星定位等。

訊源定位面臨的問題很多,例如:感測器的解析度、回音及殘響的干擾、感測器的擺放位置等等。利用不同感測器接收到訊號的時間差資訊,再做幾何定位的間接定位法,稱為訊號抵達時間差(Time Difference of Arrival, TDOA),可以解決訊號源與感測器不同步的問題,只要有足夠數量的感測器,通常都能定位到訊號源位置。但是,必須付出高成本的代價。因此,如何減少感測器數量與克服感測器的解析度限制,而能達到精準定位,是本實驗室正在努力的方向。

若是你以為單純用聲源定位就能讓宴會現場的機器人服務生聽到你的呼喚,就大錯特錯了。在雞尾酒會上,機器人身上的麥克風接收到的訊號混和多種聲音,包括:當場眾人聊天的對話聲、宴會上悠揚的鋼琴聲、宴會上碗盤的清脆響聲等等,會通通被麥克風接收成混和的聲音,這就是著名的雞尾酒會問題。雞尾酒會問題早在1953年就被提出,探討在熱鬧的雞尾派對現場有為數眾多的音源,是否能夠在派對現場吵雜的環境下,利用人造機器把某些我們想要聽到的聲音清晰地聽到或紀錄而不受其他音源干擾。想要讓機器人服務生能夠聽到你的呼喚並依指令行事,首先它必須要能夠從接收到的混和訊號中解析出你的聲音,在訊號處理的世界中,稱之為隱蔽訊號分離(Blind Source Separation, BSS)。從名稱可知,我們並不清楚混和訊號裡包含多少個訊源、也不知道包含的是哪種訊號;簡言之,就是不知道訊源的數量與種類。以一杯綜合果汁為例,包含許多不同水果,可能有香蕉、蘋果、奇異果等原料。隱蔽訊號分離就是我只有一杯綜合果汁,但是我需要分辨出內含哪些成分,更要知道這杯綜合果汁是以何種比例下去混合的。

隱蔽訊號分離最早於1985年提出。後來許多學者也跟進探討此問題並開創出許多解決方法,基本方法是建立出訊號混和的模型並分析這些訊號是如何混和的,首先要建立混和訊號的模型,以統計的方式推測訊號如何混和的初始值,接著利用演算法將混和矩陣推測出來,最後是使用混和矩陣將混和訊號的「成分」一個個分離出來。

以往隱蔽訊號分離方法多使用隨機梯度的方式,遇上不好的初始化會導致較差的分離結果,因此本實驗室嘗試用演化式機器學習得到一個合適的初始值,使得隱蔽訊號分離結果可以有高精準度。演化式機器學習是模擬生物演化設計出來的機器學習演算法,利用物競天擇的概念,逐步得到最佳的解答,而省去採用地毯式搜索需消耗的時間。常見的演化式機器學習包含基因演算法、差分演算法、粒子群演算法等。各種演化式計算有各自缺點,因此如何使用更佳的演化式機器學習演算法來得到好的初始值及隱蔽訊號分離結果,是一個研究方向。本實驗室於2011年在國際知名的演化計算期刊發表群聚引導粒子群演算法(Cluster Guide Particle Swarm Optimization, CGPSO)協助隱蔽訊號分離,就可以獲得不錯的成效。

從訊源定位、隱蔽訊號分離,到利用演化式機器學習輔助隱蔽訊號分離,我們正在逐步地解決實務上遇到的困難,相信在不久的未來,在一個宴會上你能夠呼喚遠方機器人服務生來為你服務,不會是一個僅存於電影中的場景喔!

【延伸閱讀】

  1. Tsung-Ying Sun*, Chan-Cheng Liu, Shang-Jeng Tsai, Sheng-Ta Hsieh, and Kan-Yuan Li, “Cluster Guide Particle Swarm Optimization (CGPSO) for Underdetermined Blind Source Separation with Advanced Conditions,” IEEE Transactions on Evolutionary Computation, vol. 15, no. 6, pp.798-811, Dec. 2011. (SCI, 2011 IF=3.341, 5-Year IF=4.736, Computer science: artificial intelligence, Rank=9/111)

  2. Tsung-Ying Sun*, Chan-Cheng Liu, Shang-Jeng Tsai, and Sheng-Ta Hsieh, “Blind source separation with dynamic source number using adaptive neural algorithm,” Expert Systems with Applications, vol. 36, no. 5, pp. 8855-8861, Mar. 2009. (SCI, 2009 IF=2.908, 5-Year IF=3.162, Computer science: artificial intelligence, Rank=15/103)

  3. Sheng-Ta Hsieh, Tsung-Ying Sun*, Chun-Ling Lin, and Chan-Cheng Liu, “Effective Learning Rate Adjustment of Blind Source Separation Based on an Improved Particle Swarm Optimizer,” IEEE Transactions on Evolutionary Computation, Vol. 12, No. 2, pp. 242-251, April 2008. (SCI, 2008 IF = 3.736, 5-Year IF=6.612, Computer science, artificial intelligence, Rank=3/94).

  4. Tsung-Ying Sun*, Chan-Cheng Liu, Sheng-Ta Hsieh, and Shang-Jeng Tsai, “Blind Separation with Unknown Number of Sources Based on Auto-trimmed Neural Network,” Special issue on Advanced Blind Signal Separation, Neurocomputing, vol. 71, pp. 2271-2280, Jun. 2008. (SCI, 2008 IF=1.234, 5-Year IF = 1.219, Computer science, artificial intelligence, Rank=53/94).

  5. Tian-Xi Wu, Yu-Hsiang Yu, and Tsung-Ying Sun*, “PSO-based Estimation for TDOA with Estimated Initial Values in a Reverberant Environment,” in Proc. International Conference on System Science and Engineering (ICSSE 2015), Jul. 6-8, 2015, Moriako, Japan.

  6. Tsung-Ying Sun*, Ling-Erh Lan, Chan-Cheng Liu, and Chih-Li Huo, “Mixing Matrix Identification for Underdeterminated Blind Signal Separation: Using Hough Transform and Fuzzy K-means Clustering,” in Proc. 2009 IEEE International Conference on Systems, Man and Cybernetics, pp. 1621-1626, San Antonio, Texas, USA, Oct. 14-18, 2009.

  7. Tsung-Ying Sun*, Chan-Cheng Liu, Sheng-Ta Hsieh, Tsung-Ying Tsai, and Jyun-Hong Jheng, “Optimal Determination of Wavelet Threshold and Decomposition Level via Heuristic Learning for Noise Reduction,” in Proc. 2008 IEEE Conference on Soft Computing in Industrial Applications, Muroran Hokkaido, Japan, pp. 405-410, June 25-28, 2008.

  8. Kan-Yuan Li, Tsung-Ying Sun*, Chan-Cheng Liu, Yu-Peng Jheng, Kun-Te Chien, and Chang-Ki Chou, “Density-Model-based Clustering for Blind Source Separation in Time-Frequency Domain Using PSO-FCM Hybrid Algorithm,” in Proc. 8th International Symposium on advanced Intelligent Systems, Sokcho-City, Korea, pp. 338-342, Sep. 5-8, 2007.

  9. Chan-Cheng Liu, Kan-Yuan Li, Tsung-Ying Sun*, Sheng-Ta Hsieh, and Shang-Jeng Tsai, “A Robust Blind Sparse Source Separation Algorithm Using Genetic Algorithm to Identify Mixing Matrix,” in Proc. IASTED International Conference on Signal Processing, Pattern Recognition and Applications, Innsbruck, Austria, Feb. 14-16, 2007.

  10. Chan-Cheng Liu, Tsung-Ying Sun*, Kan-Yuan Li, Sheng-Ta Hsieh, and Shang-Jeng Tsai, “Blind Sparse Source Separation Using Cluster Particle Swarm Optimization Technique,” in Proc. IASTED International Conference on Artificial Intelligence and Applications, Innsbruck, Austria, pp. 289-294, Feb. 12-14, 2007.

  11. Chan Cheng Liu, Tsung-Ying Sun*, Kan-Yuan Lee, and Chun-Ling Lin, “Underdetermined Blind Signal Separation Using Fuzzy Cluster on Mixture Accumulation,” in Proc. International Symposium on Intelligent Signal Processing and Communication Systems, Tottori, Japan, Dec. 12-15, 2006.

  12. Tsung-Ying Sun*, Chan-Cheng Liu, Sheng-Ta Hsieh, Chun-Ling Lin, and Kan-Yuan Lee, “Cluster-based Adaptive Mutation Mechanism to Improve the Performance of Genetic Algorithm.” In Proc. 6th International Conference on Intelligent Systems Design and Applications, Jinan, China, Oct. 16-18, 2006.

  13. Chan-Cheng Liu, Tsung-Ying Sun*, Sheng-Ta Hsieh, Chun-Ling Lin and Kan-Yuan Lee “A Hybrid Blind Signal Separation Algorithm: Particle Swarm Optimization on Feed-forward Neural Network,” in Proc. The 13th International Conference on Neural Information Processing, Hong Kong, Oct.3-6, 2006.

  14. Chun-Ling Lin, Sheng-Ta Hsieh, Tsung-Ying Sun*, and Chan-Cheng Liu, “PSO-Based Learning Rate Adjustment for Blind Source Separation,” in Proc. International Symposium on Intelligent Signal Processing and Communication Systems, The Chinese University of Hong Kong, Hong Kong, Dec. 13-16, 2005.

  15. Chan-Cheng Liu, Tsung-Ying Sun*, Chun-Ling Lin, and Chih-Ping Chou, “A Self-organized Neural Network for Blind Separation Process with Unobservable Sources,” in Proc. International Symposium on Intelligent Signal Processing and Communication Systems, The Chinese University of Hong Kong, Hong Kong, Dec. 13-16, 2005.

  16. Chan-Cheng Liu, Tsung-Ying Sun*, Chun-Ling Lin, and Chih-Ping Chou, “Self-Structure Feed-Forward Neural Network for Solving BSS Problem with Unknown Sources Number,” in Proc. The 12th International Conference on Neural Information Processing, Grand Hotel, Taipei, Taiwan, Oct. 30-Nov. 2, 2005.

  17. 藍翎爾、游宇翔、蔡尚錚、孫宗瀛,“隱蔽訊號分離問題-利用模糊決策機制分析非稀 疏信號成份,” 第十七屆模糊理論及其應用研討會,高雄大學,Dec. 18, 2009.

  18. 孫宗瀛*、陳美娟、劉展誠、李侃遠,“運用模糊聚類方法改善Underdetermined隱蔽稀疏信號分離效能,” 95 年度國防科技學術合作計畫成果發表會, 桃園龍潭, Nov. 30-Dec. 1, 2006.

瀏覽數