: This paper presents a novel computationally efficient voice activity detection (VAD) algorithm and emphasizes the importance of such algorithms in distributed speech recognition (DSR) systems. When ...using VAD algorithms in telecommunication systems, the required capacity of the speech transmission channel can be reduced if only the speech parts of the signal are transmitted. A similar objective can be adopted in DSR systems, where the nonspeech parameters are not sent over the transmission channel. A novel approach is proposed for VAD decisions based on mel-filter bank (MFB) outputs with the so-called Hangover criterion. Comparative tests are presented between the presented MFB VAD algorithm and three VAD algorithms used in the G.729, G.723.1, and DSR (advanced front-end) Standards. These tests were made on the Aurora 2 database, with different signal-to-noise (SNRs) ratios. In the speech recognition tests, the proposed MFB VAD outperformed all the three VAD algorithms used in the standards by
relative (G.723.1 VAD), by
relative (G.729 VAD), and by
relative (DSR VAD) in all SNRs.
A low-power, multi-stage delta-sigma modulator with comparator-based switched-capacitor (CBSC) gain stages is presented. The presented design eliminates the need for operational amplifiers and ...replaces them by comparators with current sources at their outputs to alleviate the effects of continued technology scaling on analog and mixed-signal circuits. The proposed technique significantly reduces power consumption and can be applied to switched-capacitor delta-sigma modulators of arbitrary order. Based on the proposed methodology, a 2-1 cascade, single-bit, pseudo-differential switched-capacitor delta-sigma modulator is developed and achieves a SNDR of 76.8 dB with an oversampling ratio of 64 at a clock frequency of 8 MHz.
This paper addresses the topic of defining phonetic broad classes needed during acoustic modeling for speech recognition in the procedure of decision tree based clustering. The usual approach is to ...use phonetic broad classes which are defined by an expert. This method has some disadvantages, especially in the case of multilingual speech recognition. A new data-driven method is proposed for the generation of phonetic broad classes based on a phoneme confusion matrix. The similarity measure is defined using the number of confusions between the master phoneme and all other phonemes included in the set. This proposed method is compared to the standard approach based on expert knowledge and to the randomly generated broad classes approach. The proposed data-driven method is implicitly evaluated within a speech recognition experiment. The intention of the first evaluation stage is to test the generated acoustic models in a monolingual environment (Slovenian), to show that the proposed method does not contain a multilingual influence. In the second evaluation stage, the generated acoustic models are tested in a multilingual environment (Slovenian, German and Spanish). All experiments were based on SpeechDat(II) speech databases. The proposed data-driven method for the generation of phonetic broad classes, based on phoneme confusion matrix, improved speech recognition results when compared to the method based on expert knowledge.
Modelling highly inflected languages Maučec, Mirjam Sepesy; Kačič, Zdravko; Horvat, Bogomir
Information sciences,
10/2004, Letnik:
166, Številka:
1
Journal Article
Recenzirano
Statistical language models encapsulate varied information, both grammatical and semantic, present in a language. This paper investigates various techniques for overcoming the difficulties in ...modelling highly inflected languages. The main problem is a large set of different words. We propose to model the grammatical and semantic information of words separately by splitting them into stems and endings. All the information is handled within a data-driven formalism. Grammatical information is well modelled by using short-term dependencies. This article is primarily concerned with the modelling of semantic information diffused through the entire text. It is presumed that the language being modelled is homogeneous in topic. The training corpus, which is very topically heterogeneous, is divided into three semantic levels based on topic similarity with the target environment text. Text on each semantic level is used as training text for one component of a mixture model. A document is defined as a basic unit of a training corpus, which is semantically homogeneous. The similarity of topic between a document and a collection of target environment texts is determined by the cosine vector similarity function and TFIDF weighting heuristic. The crucial question in the case of highly inflected languages is how to define terms. Terms are defined as clusters of words. Clustering is based on approximate string matching. We experimented with Levenshtein distance and Ratcliff/Obershelp similarity measure, both in combination with ending-stripping. Experiments on the Slovenian language were performed on a corpus of VEČER newswire text. The results show a significant reduction in OOV rate and perplexity.
In this paper, we propose a novel discriminative objective function for the estimation of hidden Markov model (HMM) parameters, based on the calculation of overall risk. For continuous speech ...recognition, the algorithm minimises the risk of misclassification on the training database and thus maximises recognition accuracy. The calculation of the risk is based on the measurement of Levenshtein distance between the correct transcription and the
N-best recognised transcriptions, which counts the number of recognition errors – deleted, substituted and inserted words. The minimisation of the proposed criterion is implemented by using the extended Baum–Welch algorithm for the estimation of the discrete HMM parameters and the Normandin’s extension of the algorithm for the estimation of the continuous densities. We tested the performance of the proposed algorithm on two tasks: phoneme recognition on the TIMIT database and continuous speech recognition on the Resource Management (RM) database. Results show consistent decrease of the recognition error rate when compared to the standard maximum likelihood estimation training. The highest achieved decrease of word error rate in our experiments is 20.8% on the TIMIT phoneme recognition task, 39.6% on the RM task with context-independent HMMs and 18.22% with context-dependent models.
This paper presents a novel computationally efficient voice activity detection (VAD) algorithm and emphasizes the importance of such algorithms in distributed speech recognition (DSR) systems. When ...using VAD algorithms in telecommunication systems, the required capacity of the speech transmission channel can be reduced if only the speech parts of the signal are transmitted. A similar objective can be adopted in DSR systems, where the nonspeech parameters are not sent over the transmission channel. A novel approach is proposed for VAD decisions based on mel-filter bank (MFB) outputs with the so-called Hangover criterion. Comparative tests are presented between the presented MFB VAD algorithm and three VAD algorithms used in the G.729, G.723.1, and DSR (advanced front-end) Standards. These tests were made on the Aurora 2 database, with different signal-to-noise (SNRs) ratios. In the speech recognition tests, the proposed MFB VAD outperformed all the three VAD algorithms used in the standards by 14.19% relative (G.723.1 VAD), by 12.84% relative (G.729 VAD), and by 4.17% relative (DSR VAD) in all SNRs.
This paper addresses the problem of multilingual acoustic modelling for the design of multilingual speech recognisers. An agglomerative clustering algorithm for the definition of multilingual set of ...triphones is proposed. This clustering algorithm is based on the definition of an indirect distance measure for triphones defined as a weighted sum of the explicit estimates of the context similarity on a monophone level. The monophone similarity estimation method is based on the algorithm of Houtgast. The new clustering algorithm was tested in a multilingual speech recognition experiment for three languages. The algorithm was applied on monolingual triphone sets of language specific recognisers for all languages. In order to evaluate the clustering algorithm, the performance of the multilingual set of triphones was compared to the performance of the reference system composed of all three language specific recognisers operating in parallel, and to the performance of the multilingual set of triphones produced by the tree-based clustering algorithm. All experiments were based on the 1000 FDB SpeechDat(II) databases (Slovenian, Spanish and German). Experiments have shown that the use of the clustering algorithm results in a significant reduction of the number of triphones with minor degradation of recognition rate.
Dieser Beitrag befasst sich mit dem Problem der multilingualen akustischen Modellierung für die automatische Spracherkennung. Die Verwendung eines agglomerativen Cluster-Algorithmus zur Defintion einer Menge multilingualer kontextabhängiger phonetischer Einheiten (Triphone) wird eingeführt. Der Algorithmus basiert auf einem indirekten Abstandsmaß für Triphone, das als eine gewichtete Summe der geschätzten Ähnlichkeiten der zu den Triphonen gehörenden Monophone definiert ist. Die Ähnlichkeitsschätzung der Monophone basiert auf dem Algorithmus von Houtgast. Der neue Cluster-Algorithmus wurde auf multilinguale Spracherkennungsexperimente für drei verschiedene Sprachen angewendet. Dazu wurden sprachspezifische Erkennungssysteme mit monolingualen Triphonen für alle drei Sprachen eingesetzt. Um den Cluster-Algorithmus bewerten zu können, wurde die Leistungsfähigkeit des auf den multilingualen Triphonen basierenden Systems mit zwei Referenzsystemen verglichen. Während in dem ersten Referenzsystem die sprachenspezifischen Modelle gleichzeitig (parallel) eingesetzt wurden, fanden im zweiten Referenzsystem multilinguale Modelle Verwendung, die mithilfe eines entscheidungsbaumbasierten Cluster-Algorithmus erstellt wurden. Alle Experimente basieren auf der 1000 FDB SpeechDat(II) Datenbasis (slowenisch, spanisch und deutsch). Die Untersuchungen haben gezeigt, dass die Verwendung des vorgeschlagenen agglomerativen Cluster-Algorithmus bei einer geringgradigen Abnahme der Erkennungsrate zu einer deutlichen Verringerung der Anzahl der Triphonparameter führt.
This paper presents the data-driven prediction of word level prosody breaks modeling for the Slovenian language. Automatic learning techniques depend on the construction of a large corpus labeled ...appropriately. This labeling can be done either automatically or by hand. While automatic labeling can be less accurate than hand labeling, the latter is very time consuming &, in some cases, inconsistent. Therefore, a new interactive tool for word level prosody labeling (major/minor breaks) is presented together with a new semi-automatic approach for determining prosody breaks. This interactive tool combines the advantages of hand labeling & automatic labeling by achieving a high consistency in labeling & reducing the time needed for hand labeling. The labeled Slovenian corpus has been used to train our phrase break prediction module, implementing a neural network (NN) structure. Experiments for the data-driven prediction of major = minor & major/minor phrase breaks were performed. The prediction accuracy achieved marks state-of-the-art word level prosody breaks prediction for the Slovenian language & is comparable to the prediction accuracy of other approaches in which more complex NN structures (Muller et al., 2000) or other prediction methods (Black & Tailor, 1997) were applied, & a much larger corpus was used for training. The overall prediction accuracy achieved is 94% for major = minor breaks & over 98/92% for major/minor phrase breaks, respectively. 8 Tables, 5 Figures, 21 References. Adapted from the source document
The evolution of robust speech recognition systems that maintain a high level of recognition accuracy in difficult & dynamically varying acoustical environments is becoming increasingly important as ...speech recognition technology becomes a more integral part of mobile applications. In distributed speech recognition (DSR) architecture, the recognizer's front-end is located in the terminal & is connected over a data network to a remote back-end recognition server. The terminal performs the feature parameter extraction or the front-end of the speech recognition system. These features are transmitted over a data channel to the remote back-end recognizer. DSR provides particular benefits for the applications of mobile devices such as improved recognition performance compared to using the voice channel & ubiquitous access from different networks with a guaranteed level of recognition performance. A feature extraction algorithm integrated into the DSR system is required to operate in real-time as well as with the lowest possible computational costs. In this paper, two innovative front-end processing techniques for noise robust speech recognition are presented & compared, time-domain based frame-attenuation (TD-FrAtt) & frequency-domain based frame-attenuation (FD-FrAtt). These techniques include different forms of frame-attenuation, improvement of spectral subtraction based on minimum statistics, as well as a melcepstrum feature extraction procedure. Tests are performed using the Slovenian SpeechDat II fixed telephone database & the Aurora 2 database together with the HTK speech recognition toolkit. The results obtained are especially encouraging for mobile DSR systems with limited sizes of available memory & processing power. 12 Tables, 4 Figures, 23 References. Adapted from the source document
In this paper the
harmonic features based on the harmonic decomposition of the Hildebrand–Prony line spectrum are introduced. A Hildebrand–Prony method of spectral analysis was applied because of its ...high resolution and accuracy. Comparative tests with the LP and LP-cepstral features were made with 50 speakers from the Slovene database SNABI (isolated words corpus) and 50 speakers of the German database BAS Siemens 100 (utterances of sentences). With both databases the advantages of the harmonic features were noticed especially for the speaker identification while for the speaker verification the harmonic features have performed better on the SNABI database and as good as the LP cepstral features on the BAS Siemens 100 database.
In diesem Artikel werden harmonische Eigenschaften, basierend auf einer harmonischen Dekomposition des Hildebrand–Prony Linienspektrums eingeleited. Wegen ihrer Standhaftigkeit und Genauigkeit wurde die Hildebrand–Prony-Methode verwendet. Mit 50 Sprechern aus der slowenischen Datenbank SNABI (isolierte Wortstämme) und 50 Sprechern aus der deutschen Datenbank BAS Siemens 100 (Satzaussprachen) wurden komparative Tests mit LP und LP-cepstralen Eigenschaften durchgeführt. Bei beiden Datenbank wurden Vorteile der harmonischen Eigenschaften erkannt, vor allem bei der Sprechererkennung, wobei die harmonischen Eigenschaften auf der SNABI Datenbank für die Verifizierung der Sprecher leistungsfähiger waren und auf der Datenbank BAS Siemens 100 genau so gut wie die LP-cepstrale Eigenschaften.
Cet article présente une méthode basée sur la décomposition harmonique du spectre de Hildebrand–Prony. Cette analyse spectrale de Hildebrand–Prony est appliquée pour sa haute résolution et sa précision. Des tests comparatifs avec des indices LP et LP-cepstraux ont été réalisés sur 50 sujets provenant de la base de données slovène SNABI (corpus de mots isolés) et sur 50 sujets allemands provenant de la base de données BAS Siemens 100 (phrases). Sur ces deux bases de données les avantages des indices harmonique ont été observés surtout pour l'identification des locuteurs. Pour la vérification des locuteurs, les indices harmoniques donnent des résultats meilleurs sur la base de données SNABI et aussi bons que les indices LP-cepstraux sur la base de données BAS Siemens 100.