ppt课件-a binaural processorfor robust speech recognitionin noisy（一个双耳processorfor健壮的演讲recognitionin吵了）.pptVIP

下载本文档

3
0
约1.03万字
约 38页
2017-01-25 发布于四川
举报
版权申诉

ppt课件-a binaural processorfor robust speech recognitionin noisy（一个双耳processorfor健壮的演讲recognitionin吵了）.ppt

1、原创力文档（book118）网站文档一经付费（服务费），不意味着购买了该文档的版权，仅供个人/单位学习、研究之用，不得用于商业用途，未经授权，严禁复制、发行、汇编、翻译或者网络传播等，侵权必究。。
2、本站所有内容均由合作方或网友上传，本站不对文档的完整性、权威性及其观点立场正确性做任何保证或承诺！文档内容仅供研究参考，付费前请自行鉴别。如您付费，意味着您自己接受本站规则且自行承担风险，本站不退款、不进行额外附加服务；查看《如何避免下载的几个坑》。如果您已付费下载过本站文档，您可以点击这里二次下载。
3、如文档侵犯商业秘密、侵犯著作权、侵犯人身权等，请点击“版权申诉”（推荐），也可以打举报电话：400-050-0827(电话支持时间：9:00-18:30)。
4、该文档为VIP文档，如果想要下载，成为VIP会员后，下载免费。
5、成为VIP后，下载本文档将扣除1次下载权益。下载后，不支持退款、换文档。如有疑问请联系我们。
6、成为VIP后，您将拥有八大权益，权益包括：VIP文档下载权益、阅读免打扰、文档格式转换、高级专利检索、专属身份标志、高级客服、多端互通、版权登记。
7、VIP文档为合作方或网友上传，每下载1次，网站将根据用户上传文档的质量评分、类型等，对文档贡献者给予高额补贴、流量扶持。如果你也想贡献VIP文档。上传文档

Linking Computational Auditory Scene Analysis with ‘Missing Data’ Recognition of Speech Guy J. Brown Department of Computer Science, University of Sheffield g.brown@dcs.shef.ac.uk Collaborators Kalle Palom?ki, University of Sheffield and Helsinki University of Technology DeLiang Wang, The Ohio State University Introduction Human speech perception is remarkably robust, even in the presence of interfering sounds and reverberation. In contrast, automatic speech recognition (ASR) is very problematic in such conditions: “error rates of humans are much lower than those of machines in quiet, and error rates of current recognizers increase substantially at noise levels which have little effect on human listeners” – Lippmann (1997) Can we improve ASR performance by taking an approach that models auditory processing more closely? Auditory processing in ASR Until recently, the influence of auditory processing on ASR has been largely limited to the front-end. ‘Noise robust’ feature vectors, e.g. RASTA-PLP, modulation filtered spectrograms. Can auditory processing be applied in the recogniser itself? Cooke et al. (2001) suggest that speech perception is robust because listeners can recognise speech from a partial description, i.e. with missing data. Modify conventional recogniser to deal with missing or unreliable features. Missing data approach to ASR Aim of ASR is to assign an acoustic vector Y to a class W such that the posterior probability P(W|Y) is maximised: P(W|Y) ? P(Y|W) P(W) If components of Y are unreliable or missing, cannot compute P(Y|W) as usual. Solution: partition Y into reliable parts Yr and unreliable parts Yu, and use marginal distribution P(Yr|W). Provide a time-frequency mask showing reliable regions. Missing data mask Binaural hearing and ASA Spatial location of sound sources is encoded by Interaural time difference (ITD) Interaural level difference (ILD) Spectral (pinna) cues Intelligibility of masked speech is improved if the speech and masker origin