Exceptions in language as learned by the multi-factor sparse plus low-rank language model

Language model pruning is an essential technology for speech applications running on resource-constrained devices, and many pruning algorithms have been developed for conventional word n-gram models. However, while exponential language models can give superior performance, there has been little work on the pruning of these models. In this paper, we propose several pruning algorithms for general exponential language models. We show that our best algorithm applied to an exponential n-gram model outperforms existing n-gram model pruning algorithms by up to 0.4% absolute in speech recognition word-error rate on Wall Street Journal and Broadcast News data sets. In addition, we show that Model M, an exponential class-based language model, retains its performance improvement over conventional word n-gram models when pruned to equal size, with gains of up to 2.5% absolute in word-error rate.

This paper presents two different approaches utilizing sta-tisticallanguagemodel(SLM)andsupportvectormachines(SVM) for sentence boundary detection of spontaneousJapanese. In the SLM-based approach, linguistic likeli-hoods and occurrence of pause are used to determine sen-tence boundaries. To suppress false alarms, heuristic pat-terns of end-of-sentence expressions are also incorporated.On the other hand, SVM is adopted to realize robust classi-fication against a wide variety of expressions and speechrecognition errors. Detection is performed by an SVM-based text chunker using lexical and pause information asfeatures. We evaluated these approaches on manual and au-tomatic transcription of spontaneous lectures and speeches,and achieved F-measures of 0.85 and 0.78, respectively. Index Terms : sentence boundary detection, spontaneousspeech, statistical language model, support vector ma-chines. 1. Introduction Recent advance of automatic speech recognition (ASR)technology, especially for spontaneous speech, enables var-ious applications such as spoken document archiving andretrieval, speech summarization and speech translation. Toorganizeaspokendocumentinastructuredformandtogiveuseful indices, transcriptions should be segmented into ap-propriate units like sentences. Moreover, these applicationsare usually built by combining an ASR system with naturallanguage processing (NLP) systems such as a parser and amachine translator, which often assume that input text is asentence. However,sentencesinspontaneousspeechareill-formed, and sentence boundaries are indistinct. Output textby ASR systems is just a sequence of words and has no ex-plicitsentenceboundaries, sothefurtherstepofsegmentingthe ASR output is required for these applications.Automatic boundary detection of spoken sentences hasbeen explored mainly on broadcast news (BN)tasks[1,2,3]and conversational telephone speech (CTS) tasks[3, 4, 5] inEnglish. As features for detection, pause, prosodic and lin-guistic information is often used. Most popular approach isa combination of prosodic and linguistic information[2, 3],which realizes high performance on BN and CTS tasks.Prosody-based approaches[1, 5] have also been investi-gated. Meanwhile, linguistic information is not used by it-self, since most of these works were performed on Englishdata, where cue words or expressions of sentence bound-aries are not easily defined.In Japanese, cue expressions are typically observed atthe end of sentences and expected to be useful for bound-ary detection. However, variety of such expressions is solarge in spoken Japanese, that it is hard to collect sufficientamountofdatafortrainingstatisticalmodelssuchasamax-imum entropy (ME) model. Moreover, many of cue expres-sions consist of particles, which are apparently difficult tobedetectedinASR.Thus,robustnessforASRerrorsshouldalso be investigated.In this paper, we address two approaches of sen-tence boundary detection for spontaneous Japanese. Asframeworks of detection, we adopt and compare statisti-cal language model (SLM) and support vector machines(SVM). The proposed approaches are evaluated with reallectures and speeches included in the Corpus of Sponta-neous Japanese (CSJ).

Recognition of Hungarian conversational telephone speech is challenging due to the informal style and morphological richness of the language. Recurrent Neural Network Language Model (RNNLM) can provide remedy for the high perplexity of the task; however, two-pass decoding introduces a considerable processing delay. In order to eliminate this delay we investigate approaches aiming at the complexity reduction of RNNLM, while preserving its accuracy. We compare the performance of conventional back-off n-gram language models (BNLM), BNLM approximation of RNNLMs (RNN-BNLM) and RNN n-grams in terms of perplexity and word error rate (WER). Morphological richness is often addressed by using statistically derived subwords - morphs - in the language models, hence our investigations are extended to morph-based models, as well. We found that using RNN-BNLMs 40% of the RNNLM perplexity reduction can be recovered, which is roughly equal to the performance of a RNN 4-gram model. Combining morph-based modeling and approximation of RNNLM, we were able to achieve 8% relative WER reduction and preserve real-time operation of our conversational telephone speech recognition system.

This paper introduces new methods based on exponential families for modeling the correlations between words in text and speech. While previous work assumed the effects of word co-occurrence statistics to be constant over a window of several hundred words, we show that their influence is nonstationary on a much smaller time scale. Empirical data drawn from English and Japanese text, as well as conversational speech, reveals that the "attraction" between words decays exponentially, while stylistic and syntactic contraints create a "repulsion" between words that discourages close co-occurrence. We show that these characteristics are well described by simple mixture models based on two-stage exponential distributions which can be trained using the EM algorithm. The resulting distance distributions can then be incorporated as penalizing features in an exponential language model.

This paper presents an approach to tone recognition in mandarin conversational telephone speech (CTS) based on a real context model. The real context model is proposed as a new concept designed with special consideration on the fact that mandarin CTS is characterized by complicated tone behaviors due to physiological articulation. As pitch is a supra-segmental feature, current tone's pitch value is influenced by its context especially in CTS for its fast speaking rate. A real context model covers not only the current tone but also the relative pitch level of pre-tone. Then we cluster the real context annotated training data into a few subsets to generate a more refined tone model. Gaussian Mixture Model (GMM) is used for the tone modeling. In addition, a kind of similarity measurements to compute the distance between two tones is employed, which should reveal the similarity of their pitch contour shapes and also include their different pitch height. All experiments are based on the mandarin CTS database, Train04. Our methods can improve tone recognition accuracy 4.7%.

Research in the speech recognition speech-to-text conversion area has been underway for a couple of decades, and a great deal of progress has been made in reducing the word error rate. In this paper, we attempt to summarize the state of the art in speech recognition algorithms. The algorithms we describe span the areas of lexicon design, feature extraction, classifier design, combination of hypotheses, and speaker adaptation of acoustic models. We will benchmark the algorithms on two main sources of speech, the first being Voicemail (conversational telephone speech from a single speaker) and the second being Switchboard (conversational telephone speech between two speakers). We also present the results of some cross-domain experiments which highlight the brittleness of speech recognition systems today and illustrates the need to focus research effort on improving cross-domain performance.

This paper describes the development of a speech recognition system for the processing of telephone conversations, starting with a state-of-the-art broadcast news transcription system. We identify major changes and improvements in acoustic and language modeling, as well as decoding, which are required to achieve state-of-the-art performance on conversational speech. Some major changes on the acoustic side include the use of speaker normalization (VTLN), the need to cope with channel variability, and the need for efficient speaker adaptation and better pronunciation modeling. On the linguistic side the primary challenge is to cope with the limited amount of language model training data. To address this issue we make use of a data selection technique, and a smoothing technique based on a neural network language model. At the decoding level lattice rescoring and minimum word error decoding are applied. On the development data, the improvements yield an overall word error rate of 24.9% whereas the original BN transcription system had a word error rate of about 50% on the same data.

We report insights from translating Spanish conversational telephone speech into English text by cascading an automatic speech recognition (ASR) system with a statistical machine translation (SMT) system. The key new insight is that the informal register of conversational speech is a greater challenge for ASR than for SMT: the BLEU score for translating the reference transcript is 64%, but drops to 32% for translating automatic transcripts, whose word error rate (WER) is 40%. Several strategies are examined to mitigate the impact of ASR errors on the SMT output: (i) providing the ASR lattice, instead of the 1-best output, as input to the SMT system, (ii) training the SMT system on Spanish ASR output paired with English text, instead of Spanish reference transcripts, and (iii) improving the core ASR system. Each leads to consistent and complementary improvements in the SMT output. Compared to translating the 1-best output of an ASR system with 40% WER using an SMT system trained on Spanish reference transcripts, translating the output lattice of a better ASR system with 35% WER using an SMT system trained on ASR output improves BLEU from 32% to 38%.

To use golden-angle radial sampling and compressed sensing (CS) for accelerating mono- and biexponential 3D spin-lattice relaxation time in the rotating frame (T1ρ ) mapping of knee cartilage. Golden-angle radial stack-of-stars and Cartesian 3D-T1ρ -weighted knee cartilage datasets (n = 12) were retrospectively undersampled by acceleration factors (AFs) 2-10. CS-based reconstruction using 8 different sparsifying transforms were compared for mono- and biexponential T1ρ -mapping of knee cartilage, including spatio-temporal finite differences, wavelets, dictionary from principal component analysis, and exponential decay models, and also low rank and low rank plus sparse models (L+S). Complex-valued fitting was used and Marchenko-Pastur principal component analysis filtering also tested. Most CS methods performed well for an AF of 2, with relative median normalized absolute deviation below 10% for monoexponential and biexponential mapping. For monoexponential mapping, radial sampling obtained a median normalized absolute deviation below 10% up to AF of 10, while Cartesian obtained this level of error only up to AF of 4. Radial sampling was also better with biexponential T1ρ mapping, with median normalized absolute deviation below 10% up to AF of 6. Golden-angle radial acquisitions combined with CS outperformed Cartesian acquisitions for 3D-T1ρ mapping of knee cartilage, being it is a good alternative to Cartesian sampling for reducing scan time and/or improving image and mapping quality. The methods exponential decay models, spatio-temporal finite differences, and low rank obtained the best results for radial sampling patterns.

ABSTRACTFinite memory sources and variable‐length Markov chains have recently gained popularity in data compression and mining, in particular, for applications in bioinformatics and language modelling. Here, we consider denser data compression and prediction with a family of sparse Bayesian predictive models for Markov chains in finite state spaces. Our approach lumps transition probabilities into classes composed of invariant probabilities, such that the resulting models need not have a hierarchical structure as in context tree‐based approaches. This can lead to a substantially higher rate of data compression, and such non‐hierarchical sparse models can be motivated for instance by data dependence structures existing in the bioinformatics context. We describe a Bayesian inference algorithm for learning sparse Markov models through clustering of transition probabilities. Experiments with DNA sequence and protein data show that our approach is competitive in both prediction and classification when compared with several alternative methods on the basis of variable memory length.

Discriminative dictionary learning (DDL) framework has been widely used in image classification which aims to learn some class-specific feature vectors as well as a representative dictionary according to a set of labeled training samples. However, interclass similarities and intraclass variances among input samples and learned features will generally weaken the representability of dictionary and the discrimination of feature vectors so as to degrade the classification performance. Therefore, how to explicitly represent them becomes an important issue. In this paper, we present a novel DDL framework with two-level low rank and group sparse decomposition model. In the first level, we learn a class-shared and several class-specific dictionaries, where a low rank and a group sparse regularization are, respectively, imposed on the corresponding feature matrices. In the second level, the class-specific feature matrix will be further decomposed into a low rank and a sparse matrix so that intraclass variances can be separated to concentrate the corresponding feature vectors. Extensive experimental results demonstrate the effectiveness of our model. Compared with the other state-of-the-arts on several popular image databases, our model can achieve a competitive or better performance in terms of the classification accuracy.

Maximum a posteriori (MAP) adaptation and its discriminative variants, such as MMI-MAP (maximum mutual information MAP) and MPE-MAP (minimum phone error MAP), have been widely applied to acoustic model adaptation. This paper introduces a new adaptation approach, fMPE-MAP, which is an extension to the original fMPE (feature minimum phone error) algorithm, with the enhanced ability in porting Gaussian models and fMPE transforms to a new domain. We applied this approach to the SRI-ICSI 2007 NIST meeting recognition system, for which we ported our conversational telephone speech (CTS) and broadcast news (BN) models to the meeting domain. Experiments showed that the proposed fMPE-MAP approach has comparable or better performance than simply training the fMPE transform on combined data, in addition to the obvious speed advantage. In combination with MPE-MAP, we obtained about 20% relative word error rate reduction on a lecture meeting evaluation test set, over the models trained with the standard MAP approach.

Probabilistic linear discriminant analysis (PLDA) is widely described as an effective model for text-independent speaker verification in the i-vector space. The PLDA scoring function is typically formulated as the likelihood ratio between the speaker-adapted and the universal PLDAs. In this case, the adaptation of PLDA was performed through the speaker factors. In this paper, we show that the channel factors of the PLDA could be equivalently exploited to deal with the multi-source conditions. In speaker verification, with the proposed method, a PLDAmodel trained on conversational telephone speech could be adequately adapted for interview-style microphone recordings. Experimental results on NIST SRE'08 and SRE'10 datasets confirm that the proposed method is effective, especially for the case whereby enrollment and test utterances were captured from different sources.

Phonotactic modelling, typically in the form of a PPRLM system, forms a key component in state-of-the-art Language Identification (LID) systems. Given the objective of PPRLM systems is to capture as accurately as possible the phonotactics which characterise a language, it is assumed that the minimisation of Phone Error Rate (PER) is a precursor to achieving this effectively. In this paper we examine the relevance of PER as a metric for determining eventual LID performance. In order to conduct this investigation we make use of the CallHome corpus, based on the premise it provides a better representation for the style of discourse and channel conditions encountered in the Conversational Telephone Speech (CTS), which is now the focus of current NIST LID evaluations. Using CallHome instead of the OGI-MLTS corpus to train phone recognisers, we obtained significantly improved results, with an average improvement of approximately 6% absolute across the 30, 10 and 3 seconds tasks for the NIST 1996 and 2003 evaluations. We also examine the impact of tuning the individual front-end recognisers, on both the resultant PER of other languages and against the resultant LID performance. We find that PER has a number of limitations in indicating both the degree and direction of changes to LID performance. Accordingly, we propose a new metric which is better suited for forecasting the impact on LID performance when the phone recogniser front-end is modified.

We propose a method for zero-resource domain adaptation of DNN acoustic models, for use in low-resource situations where the only in-language training data available may be poorly matched to the intended target domain. Our method uses a multi-lingual model in which several DNN layers are shared between languages. This architecture enables domain adaptation transforms learned for one well-resourced language to be applied to an entirely different low- resource language. First, to develop the technique we use English as a well-resourced language and take Spanish to mimic a low-resource language. Experiments in domain adaptation between the conversational telephone speech (CTS) domain and broadcast news (BN) domain demonstrate a 29% relative WER improvement on Spanish BN test data by using only English adaptation data. Second, we demonstrate the effectiveness of the method for low-resource languages with a poor match to the well-resourced language. Even in this scenario, the proposed method achieves relative WER improvements of 18-27% by using solely English data for domain adaptation. Compared to other related approaches based on multi-task and multi-condition training, the proposed method is able to better exploit well-resource language data for improved acoustic modelling of the low-resource target domain.