Speaker and session variability in speech processing

语音技术中的两个主要问题是说话人识别（谁在说话？）和语音识别（说了什么？）这些问题一直难以解决，因为言语中有许多类型的可变性。例如，由于通道和麦克风的影响，说话人识别系统的性能会下降，需要仔细的建模来区分这些讨厌的影响和语音信号的各个方面，这些方面有助于区分说话人。另一方面，在语音识别中，说话者之间的差异被认为是一种麻烦，而真正令人感兴趣的唯一类型的变化是语音变化。本提案涉及开发新的统计方法，使用“隐藏变量”建模扬声器和通道变化，我称之为扬声器和通道因素。我正在开发的技术纯粹是统计上的，因此很难确切地说出隐藏变量代表什么，但可以合理地推测，它们捕捉到了说话者的年龄和性别、传输通道中的电子噪声和普通背景噪声等特征。

Two of the principal problems in speech technologyare speaker recognition (who is speaking?) and speechrecognition (what is being said?). These problemshave resisted solution because of the numerous typesof variability in speech. For example, the performance ofspeaker recognition systems degrades as a resultof channel and microphone effects and careful modelingis needed to distinguish these nuisance effects from aspectsof the speech signal which are useful in discriminatingbetween speakers. In speech recognition on the other hand,variation betweenspeakers is considered to be a nuisance and the only type of variabilitywhich is of real interest is phonetic variability.This proposal is concerned with developing newstatistical methods of modeling speaker and channel variation using`hidden variables' which I call speaker and channel factors.The techniques that I am developing are purely statistical soit is hard to say exactly what the hidden variables represent butit is reasonable to surmise that they capture features such asa speaker's age and sex, electrical noise in transmission channels andordinary background acoustic noise.