基于随机轨迹模型的汉语连续语音识别方法研究

本文在指出隐马尔可夫模型（HMM）不合理假设的基础上,介绍了随机轨迹模型（STM）的理论机制及优越性。随机轨迹模型将语音基元的声学观察表示为参数空间中轨迹的聚类,并将轨迹建模为状态随机序列概率密度函数的混合,该模型可以克服HMM的不合理假设,在理论上更合理。根据STM的特点及汉语语音特色,本文对汉语连续语音识别基元的选取进行了讨论,提出了音素类单元作为识别系统的识别基元。基于STM的汉语连续语音识别的实验结果证明了STM的有效性和音素类单元的一致性。     

文本无关说话人识别的全特征矢量集模型及互信息评估方法

提出了一种文本无关说话人识别的全特征矢量集模型及互信息评估方法,该模型通过对一组说话人语音数据在特征空间进行聚类而形成,全面地反映了说话人语音的个性特征。对于说话人语音的似然度计算与判决,则提出了一种互信息评估方法,该算法综合分析距离空间和信息空间的似然度,并运用最大互信息判决准则进行识别判决。实验分析了线性预测倒谱系数(LPCC)和Mel频率倒谱系数(MFCC)两种情况下应用全特征矢量集模型和互信息评估算法的说话人识别性能,并与高斯混合模型进行了比较。结果表明:全特征矢量集模型和互信息评估算法能够充分反映说话人语音特征,并能够有效评估说话人语音特征相似程度,具有很好的识别性能,是有效的。     

一种基于音素模型感知度的发音质量评价方法

为了提高发音质量判别精度,提出了一种基于音素模型感知度的发音质量评价方法。它采用不同语音样本集合下样本声学特征的对数后验概率期望差作为音素模型对变异发音的感知度,并以此为基础,生成各音素对应的识别模型候选集。实验表明,所提出的方法使语音识别网络候选音素模型集合尺寸减少约95%;在非母语语音数据库上,该方法评分与人工专家打分相关性为0.828,基于该方法得到的声韵母错误检出率为70.8%,声调错误检出率为42.5%,均优于其它方法。      

汉语连续语音识别中一种新的音节间相关识别单元

考虑汉语连续语音中的协同发音现象对语音识别性能的提高是非常重要的。针对汉语语音的特点,提出了一种新的在汉语连续语音识别中考虑音节间协同发音现象,对声学模型进行细化的识别单元。然后基于语音学知识对音节间上下文影响进行分类,实现单元间状态参数的共享,降低了模型的复杂程度,保证了模型的可训练度。这种方法和传统方法的最大不同在于:这种方法完全利用语音学知识进行聚类,而传统方法采用数据驱动的聚类方式。识别实验表明,基于语音学分类的音节间相关识别单元对识别性能有明显的改善,系统的首选误识率降低了17％。     

面向语音增强的约束序贯高斯混合模型噪声功率谱估计

提出了一种基于极大似然的噪声对数功率谱估计方法,采用高斯混合模型对每一个频带上的功率谱包络构建统计模型,将时序包络划分为语音和非语音类,它们分别对应于高斯混合模型的两个高斯分量,描述语音和非语音的统计分布,其中非语音高斯分量的均值即为噪声功率谱的最优估计.采用序贯学习的方法,在极大似然准则下逐帧更新模型参数,并逐帧给出噪声功率谱的最优估计值。此外,由于序贯更新过程中语音信号长时缺失,容易导致模型失稳,提出了一种在线的最小描述长度准则(MDL)来判断语音信号是否长时缺失,从而保证了模型的稳定性.实验表明,算法性能整体优于经典的MS和IMCRA算法。      

汉语连续语音识别中语音处理和语言处理统合方法的研究

提出了一种语音处理和语言处理按帧同步统合的汉语连续语音识别方法。该方法把基于 ＣＦＧ语言模型和 Ｔｏｐ Ｄｏｗｎ型句法分析器的语言处理过程结合进基于有限状态自动机控制的 Ｏｎｅ Ｐａｓｓ Ｖｉｔｅｒｂｉ语音识别算法中,实现了帧同步的语音语言处理的统合。为完成帧同步句法分析的单词预测和语音识别过程的结合,本文提出了一种类似于Ｅａｒｌｅｙ法的 ＴｏｐＤｏｗｎ型句法分析方法以及 Ｏｎｅ Ｐａｓｓ Ｖｉｔｅｒｂｉ算法中的有限状态自动机动态展开建立法． ６０个音素单位和 ８个声调单位的 ＨＭＭ作为识别用基元模型被用于识别实验,识别结果表明,对于一个识别困难度（Ｐｅｒｐｌｅｘｉｔｙ）为２７．３的任务（Ｔａｓｋ）的识别系统,利用本文提出的方法,１０名话者发音的 １０７０句子的平均识别率达到 ９４．４％,比利用传统的基于单词确认（Ｗｏｒｄ Ｓｐｏｔｔｉｎｇ）以及从单词串（列）（ｌａｔｔｉｃｅ）进行句法分析的阶层性语音·语言统合方式的识别率提高约８％．     

一种机器人智能语音识别算法研究

针对智能机器人在非特定人语音识别中识别率偏低的问题,提出了一种双门限的端点检测算法,精确地检测出了语音端点,对分形维数和Mel频率倒谱系数(MFCC)进行结合,同时基于隐马尔可夫(HMM)模型,提出了智能机器人命令识别系统;在实验室环境下,利用Cool Edit软件录制了5男5女的语音,采样率为8 kHz,精度为16位,内容为5个命令词,每个词均被采集6次,将每人的前3次发音作为模板语音,后3次发音作为测试语音,实验结果表明,系统识别率可以达到85%以上,MFCC与分形维数混合的语音特征参数的算法提高了系统识别率,优化了系统性能;该方法用于非特定人语音智能识别是可行的、有效的。     

基于3维空间Viterbi算法的音素模型和声调模型识别概率统合方法的研究

提出了一种在汉语连续语音识别中基于 ３维空间 Ｖｉｔｅｒｂｉ算法的音素模型和声调模型识别概率的统合方法。该方法采用６０个音素单位的ＨＭＭ和８个声调单位的ＨＭＭ作为识别用基元模型。音素和声调基元模型识别结果的统合,采用音素的ＨＭＭ状态、声调的ＨＭＭ状态和时间的３ 维空间帧同步Ｖｉｔｅｒｂｉ 算法来实现。本文还探讨了在该方法的基础上,给予不同路径限制时的匹配统合效果,并且通过和传统的匹配统合方式的比较,证明了提出的方法的有效性。     

基于改进卷积神经网络算法的语音识别

为了解决传统卷积神经网络识别连续语音数据时识别性能较差的问题,提出一种改进的卷积神经网络算法。该方法引入Fisher准则以及L2正则化约束,在反向传播调整参数阶段,既保证参数误差的最小化,又确保分类以后的样本类间分布较分散,类内分布较集中,同时保证网络权值具有合适的数量级以有效缓解过拟合问题;采用一种更符合生物神经元激活特性的新型log激活函数进行卷积神经网络的优化,进一步提高语音识别的正确率。在语音识别库TIMIT以及THCHS30上的实验结果表明,相较于传统卷积神经网络算法,本文提出的改进算法能较好的提高语音识别率,且泛化能力更强。     

一种对加性噪声和信道函数联合补偿的模型估计方法

语音识别系统在面对实际环境中多变的加性噪声和信道差异的影响时性能急剧下降,抑制这些噪声和差异所造成的性能下降具有重要意义.作者提出了一种模型补偿算法,使用句子中的非语音段估计加性噪声,然后利用EM算法估计信道函数,从而在倒谱域上对失配的声学模型进行联合补偿.实验表明,相比基线系统,采用该算法的系统的平均性能相对提升幅度超过50%.算法可以动态跟踪环境的变化,性能表现优于一些传统的语音识别稳健性处理算法.     

Second-language experience and speech-in-noise recognition: effects of talker-listener accent similarity

Previous work has shown that the intelligibility of speech in noise is degraded if the speaker and listener differ in accent, in particular when there is a disparity between native (L1) and nonnative (L2) accents. This study investigated how this talker-listener interaction is modulated by L2 experience and accent similarity. L1 Southern British English, L1 French listeners with varying L2 English experience, and French-English bilinguals were tested on the recognition of English sentences mixed in speech-shaped noise that was spoken with a range of accents (French, Korean, Northern Irish, and Southern British English). The results demonstrated clear interactions of accent and experience, with the least experienced French speakers being most accurate with French-accented English, but more experienced listeners being most accurate with L1 Southern British English accents. An acoustic similarity metric was applied to the speech productions of the talkers and the listeners, and significant correlations were obtained between accent similarity and sentence intelligibility for pairs of individuals. Overall, the results suggest that L2 experience affects talker-listener accent interactions, altering both the intelligibility of different accents and the selectivity of accent processing.     

早晚期混响划分对理想比值掩蔽在语音识别性能上的影响

真实环境中存在的噪声和混响会降低语音识别系统的性能。封闭空间中的混响包括直达声、早期反射和后期混响3部分,它们对语音识别系统具有不同的影响.我们研究了早期反射和后期混响的不同划分方法,以其中的早期反射为目标语音,计算出了不同的理想比值掩蔽并研究了它们对语音识别系统性能的影响;在此基础上,利用双向长短时记忆网络(BLSTM)估计理想比值掩蔽,测试它们对语音识别系统性能的影响.实验结果表明,基于Abel早期反射和后期混响的划分方法,理想比值掩蔽能够降低词错误率约2.8%;基于BLSTM的估计方法过低估计了理想比值掩蔽,未能有效提高语音识别系统的性能。      

Identification and discrimination of bilingual talkers across languages

This study investigated the extent to which language familiarity affects the perception of the indexical properties of speech by testing listeners' identification and discrimination of bilingual talkers across two different languages. In one experiment, listeners were trained to identify bilingual talkers speaking in only one language and were then tested on their ability to identify the same talkers speaking in another language. In the second experiment, listeners discriminated between bilingual talkers across languages in an AX discrimination paradigm. The results of these experiments indicate that there is sufficient language-independent indexical information in speech for listeners to generalize knowledge of talkers' voices across languages and to successfully discriminate between bilingual talkers regardless of the language they are speaking. However, the results of these studies also revealed that listeners do not solely rely on language-independent information when performing these tasks. Listeners use language-dependent indexical cues to identify talkers who are speaking a familiar language. Moreover, the tendency to perceive two talkers as the "same" or "different" depends on whether the talkers are speaking in the same language. The combined results of these experiments thus suggest that indexical processing relies on both language-dependent and language-independent information in the speech signal.     

汉语自然口语中声调识别的研究

汉语是一种带声调的语言，声调信息在汉语识别中具有非常重要的意义.传统的声调识别一般只研究朗读式语音中相对标准的声调，很少对声调调型比较复杂的自然口语进行专门的处理.针对汉语自然口语的特点，在声调建模单元的选择时提出了真实上下文的模型.同时，为了对声调模式进行精细建模，采用了一种层次聚类的方法来获得更多的声调模式.实验结果证明了方法的有效性. 关键词： 声调识别 自然口语 真实上下文模型 聚类     

Effects and modeling of phonetic and acoustic confusions in accented speech

Accented speech recognition is more challenging than standard speech recognition due to the effects of phonetic and acoustic confusions. Phonetic confusion in accented speech occurs when an expected phone is pronounced as a different one, which leads to erroneous recognition. Acoustic confusion occurs when the pronounced phone is found to lie acoustically between two baseform models and can be equally recognized as either one. We propose that it is necessary to analyze and model these confusions separately in order to improve accented speech recognition without degrading standard speech recognition. Since low phonetic confusion units in accented speech do not give rise to automatic speech recognition errors, we focus on analyzing and reducing phonetic and acoustic confusability under high phonetic confusion conditions. We propose using likelihood ratio test to measure phonetic confusion, and asymmetric acoustic distance to measure acoustic confusion. Only accent-specific phonetic units with low acoustic confusion are used in an augmented pronunciation dictionary, while phonetic units with high acoustic confusion are reconstructed using decision tree merging. Experimental results show that our approach is effective and superior to methods modeling phonetic confusion or acoustic confusion alone in accented speech, with a significant 5.7% absolute WER reduction, without degrading standard speech recognition.     

Evidence against the mismatched interlanguage speech intelligibility benefit hypothesis

In a follow-up study to that of Bent and Bradlow (2003), carrier sentences containing familiar keywords were read aloud by five talkers (Korean high proficiency; Korean low proficiency; Saudi Arabian high proficiency; Saudi Arabian low proficiency; native English). The intelligibility of these keywords to 50 listeners in four first language groups (Korean, n = 10; Saudi Arabian, n = 10; native English, n = 10; other mixed first languages, n = 20) was measured in a word recognition test. In each case, the non-native listeners found the non-native low-proficiency talkers who did not share the same first language as the listeners the least intelligible, at statistically significant levels, while not finding the low-proficiency talker who shared their own first language similarly unintelligible. These findings indicate a mismatched interlanguage speech intelligibility detriment for low-proficiency non-native speakers and a potential intelligibility problem between mismatched first language low-proficiency speakers unfamiliar with each others' accents in English. There was no strong evidence to support either an intelligibility benefit for the high-proficiency non-native talkers to the listeners from a different first language background or to indicate that the native talkers were more intelligible than the high-proficiency non-native talkers to any of the listeners.     

An effective cluster-based model for robust speech detection and speech recognition in noisy environments

This paper shows an accurate speech detection algorithm for improving the performance of speech recognition systems working in noisy environments. The proposed method is based on a hard decision clustering approach where a set of prototypes is used to characterize the noisy channel. Detecting the presence of speech is enabled by a decision rule formulated in terms of an averaged distance between the observation vector and a cluster-based noise model. The algorithm benefits from using contextual information, a strategy that considers not only a single speech frame but also a neighborhood of data in order to smooth the decision function and improve speech detection robustness. The proposed scheme exhibits reduced computational cost making it adequate for real time applications, i.e., automated speech recognition systems. An exhaustive analysis is conducted on the AURORA 2 and AURORA 3 databases in order to assess the performance of the algorithm and to compare it to existing standard voice activity detection (VAD) methods. The results show significant improvements in detection accuracy and speech recognition rate over standard VADs such as ITU-T G.729, ETSI GSM AMR, and ETSI AFE for distributed speech recognition and a representative set of recently reported VAD algorithms.     

基于Transformer的普通话语声识别模型位置编码选择

具有自注意机制的Transformer网络在语声识别研究领域渐渐得到广泛关注。该文围绕着将位置信息嵌入与语声特征相结合的方向,研究更加适合普通话语声识别模型的位置编码方法。实验结果得出,采用卷积编码的输入表示代替正弦位置编码,可以更好地融合语声特征上下文联系和相对位置信息,获得较好的识别效果。训练的语声识别系统是在Transformer模型基础上,比较4种不同的位置编码方法。结合3-gram语言模型,所提出的卷积位置编码方法,在中文语声数据集AISHELL-1上的误识率降低至8.16%。     

Effects of cross-language voice training on speech perception: whose familiar voices are more intelligible?

Previous research has shown that familiarity with a talker's voice can improve linguistic processing (herein, "Familiar Talker Advantage"), but this benefit is constrained by the context in which the talker's voice is familiar. The current study examined how familiarity affects intelligibility by manipulating the type of talker information available to listeners. One group of listeners learned to identify bilingual talkers' voices from English words, where they learned language-specific talker information. A second group of listeners learned the same talkers from German words, and thus only learned language-independent talker information. After voice training, both groups of listeners completed a word recognition task with English words produced by both familiar and unfamiliar talkers. Results revealed that English-trained listeners perceived more phonemes correct for familiar than unfamiliar talkers, while German-trained listeners did not show improved intelligibility for familiar talkers. The absence of a processing advantage in speech intelligibility for the German-trained listeners demonstrates limitations on the Familiar Talker Advantage, which crucially depends on the language context in which the talkers' voices were learned; knowledge of how a talker produces linguistically relevant contrasts in a particular language is necessary to increase speech intelligibility for words produced by familiar talkers.