Android平台下OpenCL加速的说话人识别系统
|
摘要
如今,人工智能正在图像、自然语言处理等诸多领域迅速发展,同时随着移动设备的广泛使用,人们的生活习惯正在逐步的改变。所以,将人工智能技术运用到移动互联网中已经成为必然趋势。但由于移动设备因密集的计算带来的功耗提升和存储带宽的增加,使得在移动设备中实现神经网络算法面临着巨大的挑战。说话人识别技术作为一种安全可靠的生物认证技术,将其运用到移动设备平台中有着其他生物认证技术没有的便捷性和安全性,同时为了提高效率,论文提出在Android平台下,通过提取说话人的梅尔倒谱特征(MFCC),使用OpenCL对基于BP神经网络的说话人识别系统进行加速,通过实验对比加速前后的运行效率,可以发现在Android平台下使用OpenCL加速,可以提升计算速度。
Nowadays,with the rapidly development of Artificial Intelligence in many areas such as images and natural language processing,and the widespread usage of mobile devices,the living habits of people are changing gradually. Therefore,it has become an inexorable trend of applying the artificial intelligence technology to mobile internet. However,due to the increasing of power dissipation and storage bandwidth caused by intensely computing,the application of the neural network algorithm in mobile devices is facing a great challenge. As a safe and reliable biometric authentication,the applying of speakers identification technology in speakers identification technology has the unique advantages of convenience and safety. This article proposed that in the Android platform,by extracting the speaker's Mel-Frequency Cepstral Coefficients(MFCC),using OpenCL to accelerate the speaker identification system based on bp neural network,comparing the efficiency before and after the accelerating via the experiment,it is found that the computation speed can be improved by using OpenCL on the Android platform.
Nowadays,with the rapidly development of Artificial Intelligence in many areas such as images and natural language processing,and the widespread usage of mobile devices,the living habits of people are changing gradually. Therefore,it has become an inexorable trend of applying the artificial intelligence technology to mobile internet. However,due to the increasing of power dissipation and storage bandwidth caused by intensely computing,the application of the neural network algorithm in mobile devices is facing a great challenge. As a safe and reliable biometric authentication,the applying of speakers identification technology in speakers identification technology has the unique advantages of convenience and safety. This article proposed that in the Android platform,by extracting the speaker's Mel-Frequency Cepstral Coefficients(MFCC),using OpenCL to accelerate the speaker identification system based on bp neural network,comparing the efficiency before and after the accelerating via the experiment,it is found that the computation speed can be improved by using OpenCL on the Android platform.
引文
[1]Xu J Y. OpenCL-The Open Standard for Parallel Programming of Heterogeneous Systems[J]. 2011.
[2]Ross J A,Richie D A,Song J P,et al. A case study of OpenCL on an Android mobile GPU[C]//High PERFORMANCE Extreme Computing Conference. IEEE,2015:1-6.
[3]Campbell J P J. Speaker recognition:a tutorial[J]. Proceedings of the IEEE,1997,85(9):1437-1462.
[4]Gish H,Schmidt M. Text-independent speaker identification[J]. IEEE Signal Processing Magazine,1994,11(4):18-32.
[5] Kinnunen T,Li H. An overview of text-independent speaker recognition:From features to supervectors[J].Speech Communication,2010,52(1):12-40.
[6]赵力.语音信号处理-第2版[M].北京:机械工业出版社,2009.ZHAO Li. Speech signal processing-version 2[M]. BeiJing:Mechanical industry press,2009.
[7]Siniscalchi S M,Yu D,Deng L,et al. Speech Recognition Using Long-Span Temporal Patterns in a Deep Network Model[J]. IEEE Signal Processing Letters,2013,20(3):201-204.
[8]刘加.汉语大词汇量连续语音识别系统研究进展[J].电子学报,2000,28(1):85-91.Research on Large Vocabulary Continuous Speech Recognition System[J]. Acta Electronica Sinica,2000,28(1):85-91.
[9]Kinnunen T,Li H. An overview of text-independent speaker recognition:From features to supervectors[M]. Elsevier Science Publishers B. V. 2010.
[10]Oglesby J,Mason J S. Optimisation of Neural Models for Speaker Identification[J]. 1990,1:261-264.
[11]张徽强,谭吉春,王晓峰.语音信号中字词端点检测方法的两项改进[J].电光与控制,2005,12(4):68-70.ZHANG Huiqiang,TAN Jichun,WANG Xiaofeng. Improvement on word endpoint detection in phonetic signal processing[J]. Electronics Optics&Control,2005,12(4):68-70.
[12]Chawla N V,Bowyer K W,Hall L O,et al. SMOTE:synthetic minority over-sampling technique[J]. Journal of Artificial Intelligence Research,2002,16(1):321-357.
[13]The OpenCL Specificatin Version:2.0. Khronos OpenCL Working Group,2015.
[14]Wang B,Zhu L,Jia K,et al. Accelerated Cone Beam CT Reconstruction Based on OpenCL[C]//2010 international conference on image analysis and signal processing,2010:291-295.
[15]Robert C. Machine Learning,a Probabilistic Perspective[J]. Mathematics Education Library,2014,58(8):27-71.
[2]Ross J A,Richie D A,Song J P,et al. A case study of OpenCL on an Android mobile GPU[C]//High PERFORMANCE Extreme Computing Conference. IEEE,2015:1-6.
[3]Campbell J P J. Speaker recognition:a tutorial[J]. Proceedings of the IEEE,1997,85(9):1437-1462.
[4]Gish H,Schmidt M. Text-independent speaker identification[J]. IEEE Signal Processing Magazine,1994,11(4):18-32.
[5] Kinnunen T,Li H. An overview of text-independent speaker recognition:From features to supervectors[J].Speech Communication,2010,52(1):12-40.
[6]赵力.语音信号处理-第2版[M].北京:机械工业出版社,2009.ZHAO Li. Speech signal processing-version 2[M]. BeiJing:Mechanical industry press,2009.
[7]Siniscalchi S M,Yu D,Deng L,et al. Speech Recognition Using Long-Span Temporal Patterns in a Deep Network Model[J]. IEEE Signal Processing Letters,2013,20(3):201-204.
[8]刘加.汉语大词汇量连续语音识别系统研究进展[J].电子学报,2000,28(1):85-91.Research on Large Vocabulary Continuous Speech Recognition System[J]. Acta Electronica Sinica,2000,28(1):85-91.
[9]Kinnunen T,Li H. An overview of text-independent speaker recognition:From features to supervectors[M]. Elsevier Science Publishers B. V. 2010.
[10]Oglesby J,Mason J S. Optimisation of Neural Models for Speaker Identification[J]. 1990,1:261-264.
[11]张徽强,谭吉春,王晓峰.语音信号中字词端点检测方法的两项改进[J].电光与控制,2005,12(4):68-70.ZHANG Huiqiang,TAN Jichun,WANG Xiaofeng. Improvement on word endpoint detection in phonetic signal processing[J]. Electronics Optics&Control,2005,12(4):68-70.
[12]Chawla N V,Bowyer K W,Hall L O,et al. SMOTE:synthetic minority over-sampling technique[J]. Journal of Artificial Intelligence Research,2002,16(1):321-357.
[13]The OpenCL Specificatin Version:2.0. Khronos OpenCL Working Group,2015.
[14]Wang B,Zhu L,Jia K,et al. Accelerated Cone Beam CT Reconstruction Based on OpenCL[C]//2010 international conference on image analysis and signal processing,2010:291-295.
[15]Robert C. Machine Learning,a Probabilistic Perspective[J]. Mathematics Education Library,2014,58(8):27-71.