第一讲-语音识别概述.pptVIP

语音识别简介 语音识别 语音识别研究机构 语音识别的应用 语音信号处理的范畴 语音信号处理的研究方向 商业技术报告 语音识别 ASR的定义 ASR的流程 ASR的分类 ASR的学科特点 ASR的历史 ASR的定义 两个不同层次的理解 ASR:Automatic Speech Recognition，让机器把人说的语音翻译成文字 自动语音理解（ASU）:在语音识别的基础上，理解人所说的文字意义 目标：理解任何人在任何环境下的语音，模仿人的听觉能力 ASR的信号流框图 Front-end analysis 语音识别分类 按词汇量大小 小词汇量（30）、中词汇量(100) 大词汇量(无限量) 发音是否连续：孤立词、连续语音 按与说话人的关系：SD，SI 按说话方式：朗诵语音、口语语音 按应用目标：通用、专用 按实现方法：计算机软件，嵌入式系统，计算机网络，电话网络，无线网络 学科特点:交叉学科 Signal Processing Acoustics Pattern recognition Communication and information theory Linguistics Physiology and psychology Computer Science and technology Cognitive science History What was the first success story of speech recognition? “Radio Rex” in the 1922, was the first success story in the field of speech recognition little toy dog named Rex” Voice recognition History 1936 - AT T’s Bell labs started study of speech recognition (funded by DARPA) 1950s-Isolated digit,syllables, phonemes, small vocabulary 1960s Japan, Hardware systems Use of dynamic programming Reddy (CMU) :continuous speech recognition History(con’t) 1970s, significant milestones Isolated word became viable and usable: using pattern-recognition, DP and LPC IBM’s efforts in large vocabulary ATT Bell Labs’s efforts in SI ASR, -clustering algorithms 1978 – speak and spell toy by TI History(con’t) 1980s, Focus on connected word recognition two-level DP (NEC),One-pass method,Level building Shift from template-based approaches to statistical modeling methods –HMM Applying ANN in the late 1980s Systems: SPHINX (CMU), BYBLOS (BBN), SRI (Lincoln Labs) History(con’t) 1990s, Focus on continuous ASR Reduce the effect of noise Robustness On-line learning (Speaker adaptation) Natural language front ends to ASR IBM Via Voice Development platform：VBVoice (Pronexus) 1997 – Dragon Systems History(con’t) 21c: Technology to deal with environment variation ASR for multi-languages Embedded System Challenges of speech recognition Ease of use Robust performance Automatic learning of new words and sounds Grammar for spo