Improving Phoneme Classification Performance Using Observation Context–Dependent Segment Models

This article describes a novel method that models the correlation among acoustic observations in contiguous speech segments. The basic idea behind the method is that acoustic observations are conditioned not only on the phonetic context but also on the preceding acoustic segment observation. The correlation between consecutive acoustic observations is modeled by mean trajectory polynomial segment models (PSM). This method is an extension of conventional segment modeling approaches in that it describes the correlation of acoustic observations not only inside segments but also between contiguous segments. It is also a generalization of phonetic context (e.g., triphone) modeling approaches because it can model acoustic context and phonetic context at the same time. Using the proposed method in a speaker-independent phoneme classification test resulted in a 7 to 9% relative reduction of error rate as compared with the traditional triphone segmental model system and a 31% reduction as compared with a similar triphone hidden Markov model (HMM) system.     

汉语语音识别中声学界标点引导的随机段模型解码算法

提出了一种随机段模型的解码优化算法.检测出具有语音学意义的界标点,根据这些界标点分析临近语音段的边界信息和声韵母类别信息,最后将这些边界信息和类别信息用于指导随机段模型的搜索过程.实验中,两种类型的界标点能较为准确地被检测出来,并用于指导随机段模型的解码,在“863-test”测试集上进行的汉语连续语音识别实验显示,在正确率只有轻微下降的同时,解码时间下降了12.92％,这表明了将语音学知识引入语音识别系统的有效性.     

一种新的基于并行分段裁剪的DTW算法

通过对DTW算法的研究,提出了一种并行分段裁剪的新方法,在减少DTW算法运算量方面有显著效果,并将其用于一个地名识别系统中,经测试,可以明显缩短识别时间,具有很强的实时性,有较高的识别率,适合作为小型语音识别产品的主要算法。     

Combining missing-feature theory, speech enhancement, and speaker-dependent/-independent modeling for speech separation

This paper considers the separation and recognition of overlapped speech sentences assuming single-channel observation. A system based on a combination of several different techniques is proposed. The system uses a missing-feature approach for improving crosstalk/noise robustness, a Wiener filter for speech enhancement, hidden Markov models for speech reconstruction, and speaker-dependent/-independent modeling for speaker and speech recognition. We develop the system on the Speech Separation Challenge database, involving a task of separating and recognizing two mixing sentences without assuming advanced knowledge about the identity of the speakers nor about the signal-to-noise ratio. The paper is an extended version of a previous conference paper submitted for the challenge.     

语音识别中基于发音特征的声调集成算法

提出基于发音特征的声调建模改进方法,并将其用于随机段模型的一遍解码中。根据普通话的发音特点,确定了用于区别汉语元音、辅音信息的7种发音特征,并以此为目标值利用阶层式多层感知器计算语音信号属于发音特征的35个类别后验概率,将该概率作为发音特征与传统的韵律特征一起用于声调建模。根据随机段模型的解码特点,在两层剪枝后对保留下来的路径计算其声调模型概率得分,加权后加入路径总的概率得分中。在“863-test”测试集上进行的实验结果显示,使用了新的发音特征集合中声调模型的识别精度提高了3.11%;融入声调信息后随机段模型的字错误率从13.67%下降到12.74%。表明了将声调信息应用到随机段模型的可行性。     

火车票查询系统中语音识别的研究及实现

文章首先介绍了火车票查询系统中语音识别的框架结构,并详细描述了采用微软SPEECHSDK技术实现车次、车站语音识别的详细流程,最后从识别率、鲁棒性方面对该识别系统进行测试和分析,实验表明,该语音识别系统是稳定的和实用的。     

Spectral domain texture analysis for speech enhancement

A fast algorithm which aims at performing texture analysis of time-frequency images for denoising purposes is described in this paper. Time-frequency images are built using the peaks of the amplitude spectrum computed on a noisy speech signal. Using texture analysis, we can look at the spectral 2D information on a large scale, thus allowing the correction of spectral continuity by restoring peaks corrupted by noise which can appear as missing or modified. The algorithm has been used in preprocessors of speech processing systems. In fact, we report interesting results obtained with this algorithm in speech enhancement and HMM speech recognition tasks, especially for noise types which are quite difficult to treat with conventional algorithms, such as micro-interruptions or bursts of tonal noise at random frequencies.     

深度语音信号与信息处理：研究进展与展望

论文首先对深度学习进行简要的介绍,然后就其在语音信号与信息处理研究领域的主要研究方向,包括语音识别、语音合成、语音增强的研究进展进行了详细的介绍。语音识别方向主要介绍了基于深度神经网络的语音声学建模、大数据下的模型训练和说话人自适应技术;语音合成方向主要介绍了基于深度学习模型的若干语音合成方法;语音增强方向主要介绍了基于深度神经网络的若干典型语音增强方案。论文的最后我们对深度学习在语音信与信息处理领域的未来可能的研究热点进行展望。     

Breadth-first search strategies for trie-based syntactic pattern recognition

Dictionary-based syntactic pattern recognition of strings attempts to recognize a transmitted string X ^*, by processing its noisy version, Y, without sequentially comparing Y with every element X in the finite, (but possibly, large) dictionary, H. The best estimate X ⁺ of X ^*, is defined as that element of H which minimizes the generalized Levenshtein distance (GLD) D(X, Y) between X and Y, for all X ∈H. The non-sequential PR computation of X ⁺ involves a compact trie-based representation of H. In this paper, we show how we can optimize this computation by incorporating breadth first search schemes on the underlying graph structure. This heuristic emerges from the trie-based dynamic programming recursive equations, which can be effectively implemented using a new data structure called the linked list of prefixes that can be built separately or “on top of” the trie representation of H. The new scheme does not restrict the number of errors in Y to be merely a small constant, as is done in most of the available methods. The main contribution is that our new approach can be used for generalized GLDs and not merely for 0/1 costs. It is also applicable when all possible correct candidates need to be known, and not just the best match. These constitute the cases when the “cutoffs” cannot be used in the DFS trie-based technique (Shang and Merrettal in IEEE Trans Knowl Data Eng 8(4):540–547, 1996). The new technique is compared with the DFS trie-based technique (Risvik in United Patent 6377945 B1, 23 April 2002; Shang and Merrettal in IEEE Trans Knowl Data Eng 8(4):540–547, 1996) using three large and small benchmark dictionaries with different errors. In each case, we demonstrate marked improvements with regard to the operations needed up to 21%, while at the same time maintaining the same accuracy. Additionally, some further improvements can be obtained by introducing the knowledge of the maximum number or percentage of errors in Y.

面向情感语音识别的建模方法研究

情感是语音识别研究中一个不可避免的问题,不同的情感对于语音有着不同的影响,这种影响使得中性语音识别系统在实际应用中的识别效果大打折扣。对于类似的影响通常的解决方法有寻找鲁棒特征,特征归一化以度模型调整训练等。本文通过自适应方法,使用少量情感数据,在中性语音模型的基础上自适应得到新的情感语音模型。实验证明,新模型对于情感语音有着更好的整体识别率。     

基于改进MFCC特征提取和DNN网络的机器人语音识别方法研究

为了实现机器人语音控制,并避免环境噪音的干扰,研究提出了基于Mel频率倒谱系数特征提取和深度神经网络的机器人语音控制指令识别方法。实验结果显示,相较于其他语音增强方法,基于深度神经网络和谐波增强技术的语音增强方法分段信噪比和语音质量感观评价均更高。同时相比于其他特征,研究提出的基于改进Mel频率倒谱系数特征能显著降低语音识别的字错误率,通过辅以改进深度神经网络-隐马尔科夫模型能进一步降低字错误率。在20dB条件下,该特征和改进深度神经网络-隐马尔科夫模型的平均字错误率分别为24.9%和22.1%,均低于其他方法。上述结果表明,研究提出的语音识别方法能实现带噪声语音的准确识别,提高机器人的语音控制指令识别能力。     

带噪语音处理系统的研究与设计

介绍利用多媒体处理技术,在多媒体计算机上实现的从带噪声的语音信号中提取原始语音方法的研究及该方法的设计与开发。     

Optimal algorithms for locating the longest and shortest segments satisfying a sum or an average constraint

We study several fundamental problems arising from biological sequence analysis. Given a sequence of real numbers, we present two linear-time algorithms, one for locating the “longest” sum-constrained segment, and the other for locating the “shortest” sum-constrained segment. These two algorithms are based on the same framework and run in an online manner, hence they are capable of handling data stream inputs. Our algorithms also yield online linear-time solutions for finding the longest and shortest average-constrained segments by a simple reduction.     

一种新的工业机器人语音增强方法设计

分析和研究了基于声波耦合和语音增强模块级联的语音增强方法的工业语音识别系统设计和实施过程,并对其进行了算法建模,同时在比较谱减法和MMSE-LSA的语音增强算法的同时进行了实验数据分析,使工业机器人语音识别系统在噪声环境下提高了识别率.     

噪声鲁棒语音识别研究综述

针对噪声环境下的语音识别问题,对现有的噪声鲁棒语音识别技术进行讨论,阐述了噪声鲁棒语音识别研究的主要问题,并根据语音识别系统的构成将噪声鲁棒语音识别技术按照信号空间、特征空间和模型空间进行分类总结,分析了各种鲁棒语音识别技术的特点、实现,以及在语音识别中的应用。最后展望了进一步的研究方向。     

语音识别技术分析

语音识别是研究让机器能够听懂人类口述的自然语言的一门学科,其最终目标是实现入与机器进行自然语言通信。介绍了语音识别的关键技术,主要有语音识别单元选取、特征参数提取技术、模式匹配准则及模型训练技术三个方面,通过具体的实现方法对各种技术作出介绍,以达到在语音识别中提高识别率的效果。最后对各种技术以及语音识别系统的作用和意义作出简要总结。     

An adaptive post-filtering method producing an artificial Lombard-like effect for intelligibility enhancement of narrowband telephone speech

Post-filtering can be used in mobile communications to improve the quality and intelligibility of speech. Energy reallocation with a high-pass type filter has been shown to work effectively in improving the intelligibility of speech in difficult noise conditions. This paper introduces a post-filtering algorithm that adapts to the background noise level as well as to the fundamental frequency of the speaker and models the spectral effects observed in natural Lombard speech. The introduced method and another post-filtering technique were compared to unprocessed telephone speech in subjective listening tests in terms of intelligibility and quality. The results indicate that the proposed method outperforms the reference method in difficult noise conditions.     

智能电话查询系统

介绍了一个用于交通信息查询的交互式对话系统-ShanghaiQuest。该系统集成了语音识别、文语转换和自然语言理解等技术，用户可以同这样一个自动的交通代理对话以获取所需的交通信息。