Phone Recognition Research Articles

Advanced classification approach for neuronal phoneme recognition system based on efficient constructive training algorithm

This paper introduces a neural network optimization procedure allowing the generation of multilayer perceptron (MLP) network topologies with few connections, low complexity and high classification performance for phoneme's recognition. An efficient constructive algorithm with incremental training using a new proposed Frame by Frame Neural Networks (FFNN) classification approach for automatic phoneme recognition is thus proposed. It is based on a novel recruiting hidden neuron's procedure for a single hidden-layer. After an initializing phase started with initial small number of hidden neurons, this algorithm allows the Neural Networks (NNs) to adjust automatically its parameters during the training phase. The modular FFNN classification method is then constructed and tested to recognize 5 broad phonetic classes extracted from the TIMIT database. In order to take into account the speech variability related to the coarticulation effect, a Context Window of Three Successive Frame's (CWTSF) analysis is applied. Although, an important reduction of the computational training time is observed, this technique penalized the overall Phone Recognition Rate (PRR) and increased the complexity of the recognition system. To alleviate these limitations, two feature dimensionality reduction techniques respectively based on Principal Component Analysis (PCA) and Self Organizing Maps (SOM) are investigated. It is observed an important improvement in the performance of the recognition system when the PCA technique is applied. Optimal neuronal phone recognition architecture is finally derived according to the following criteria: best PRR, minimum computational training time and complexity of the BPNN architecture.

Phoneme recognition using zerocrossing interval distribution of speech patterns and ANN

The speech signal is modeled using zerocrossing interval distribution of the signal in time domain. The distributions of these parameters are studied over five Malayalam (one of the most popular Indian language) vowels. We found that the distribution patterns are almost similar for repeated utterances of the same vowel and varies from vowel to vowel. These distribution patterns are used for recognizing the vowels using multilayer feed forward artificial neural network. After analyzing the distribution patterns and the vowel recognition results, we realize that the zerocrossing interval distribution parameters can be effectively used for the speech phone classification and recognition. The noise adaptness of this parameter is also studied by adding additive white Gaussian noise at different signal to noise ratio. The computational complexity of the proposed technique is also less compared to the conventional spectral techniques, which includes FFT and Cepstral methods, used in the parameterization of speech signal.

Broad phonetic class definition driven by phone confusions

Intermediate representations between the speech signal and phones may be used to improve discrimination among phones that are often confused. These representations are usually found according to broad phonetic classes, which are defined by a phonetician. This article proposes an alternative data-driven method to generate these classes. Phone confusion information from the analysis of the output of a phone recognition system is used to find clusters at high risk of mutual confusion. A metric is defined to compute the distance between phones. The results, using TIMIT data, show that the proposed confusion-driven phone clustering method is an attractive alternative to the approaches based on human knowledge. A hierarchical classification structure to improve phone recognition is also proposed using a discriminative weight training method. Experiments show improvements in phone recognition on the TIMIT database compared to a baseline system.

Automatic speech recognition performance in different room acoustic environments with and without dereverberation preprocessing

The performance of recent dereverberation methods for reverberant speech preprocessing prior to Automatic Speech Recognition (ASR) is compared for an extensive range of room and source-receiver configurations. It is shown that room acoustic parameters such as the clarity (C50) and the definition (D50) correlate well with the ASR results. When available, such room acoustic parameters can provide insight into reverberant speech ASR performance and potential improvement via dereverberation preprocessing. It is also shown that the application of a recent dereverberation method based on perceptual modelling can be used in the above context and achieve significant Phone Recognition (PR) improvement, especially under highly reverberant conditions.

Experiments on Cross-Language Attribute Detection and Phone Recognition With Minimal Target-Specific Training Data

A state-of-the-art automatic speech recognition (ASR) system can often achieve high accuracy for most spoken languages of interest if a large amount of speech material can be collected and used to train a set of language-specific acoustic phone models. However, designing good ASR systems with little or no language-specific speech data for resource-limited languages is still a challenging research topic. As a consequence, there has been an increasing interest in exploring knowledge sharing among a large number of languages so that a universal set of acoustic phone units can be defined to work for multiple or even for all languages. This work aims at demonstrating that a recently proposed automatic speech attribute transcription framework can play a key role in designing language-universal acoustic models by sharing speech units among all target languages at the acoustic phonetic attribute level. The language-universal acoustic models are evaluated through phone recognition. It will be shown that good cross-language attribute detection and continuous phone recognition performance can be accomplished for “unseen” languages using minimal training data from the target languages to be recognized. Furthermore, a phone-based background model (PBM) approach will be presented to improve attribute detection accuracies.

Multi-accent acoustic modelling of South African English

Although English is spoken throughout South Africa it is most often used as a second or third language, resulting in several prevalent accents within the same population. When dealing with multiple accents in this under-resourced environment, automatic speech recognition (ASR) is complicated by the need to compile multiple, accent-specific speech corpora. We investigate how best to combine speech data from five South African accents of English in order to improve overall speech recognition performance. Three acoustic modelling approaches are considered: separate accent-specific models, accent-independent models obtained by pooling training data across accents, and multi-accent models. The latter approach extends the decision-tree clustering process normally used to construct tied-state hidden Markov models (HMMs) by allowing questions relating to accent. We find that multi-accent modelling outperforms accent-specific and accent-independent modelling in both phone and word recognition experiments, and that these improvements are statistically significant. Furthermore, we find that the relative merits of the accent-independent and accent-specific approaches depend on the particular accents involved. Multi-accent modelling therefore offers a mechanism by which speech recognition performance can be optimised automatically, and for hard decisions regarding which data to pool and which to separate to be avoided.

Acoustic Modeling Using Deep Belief Networks

Gaussian mixture models are currently the dominant technique for modeling the emission distribution of hidden Markov models for speech recognition. We show that better phone recognition on the TIMIT dataset can be achieved by replacing Gaussian mixture models by deep neural networks that contain many layers of features and a very large number of parameters. These networks are first pre-trained as a multi-layer generative model of a window of spectral feature vectors without making use of any discriminative information. Once the generative pre-training has designed the features, we perform discriminative fine-tuning using backpropagation to adjust the features slightly to make them better at predicting a probability distribution over the states of monophone hidden Markov models.

Context-Dependent Pre-Trained Deep Neural Networks for Large-Vocabulary Speech Recognition

We propose a novel context-dependent (CD) model for large-vocabulary speech recognition (LVSR) that leverages recent advances in using deep belief networks for phone recognition. We describe a pre-trained deep neural network hidden Markov model (DNN-HMM) hybrid architecture that trains the DNN to produce a distribution over senones (tied triphone states) as its output. The deep belief network pre-training algorithm is a robust and often helpful way to initialize deep neural networks generatively that can aid in optimization and reduce generalization error. We illustrate the key components of our model, describe the procedure for applying CD-DNN-HMMs to LVSR, and analyze the effects of various modeling choices on performance. Experiments on a challenging business search dataset demonstrate that CD-DNN-HMMs can significantly outperform the conventional context-dependent Gaussian mixture model (GMM)-HMMs, with an absolute sentence accuracy improvement of 5.8% and 9.2% (or relative error reduction of 16.0% and 23.2%) over the CD-GMM-HMMs trained using the minimum phone error rate (MPE) and maximum-likelihood (ML) criteria, respectively.

Combining formant frequency based on variable order LPC coding with acoustic features for TIMIT phone recognition

Combination of multiple acoustic features has great potential to improve Automatic Speech Recognition (ASR) accuracy. Our contribution in this research was to investigate one novel method when using voiced formants' features in combination with standard MFCC features in order to enhance TIMIT phone recognition. These voiced features provide accurate formants frequencies using a Variable Order LPC Coding (VO-LPC) algorithm that was combined with continuity constraints. The overall estimating formants were concatenated with MFCC features when a voiced frame could be detected. For feature-level combination, Heteroscedastic Linear Discriminant Analysis (HLDA) based approach had been used successfully to find an optimal linear combination of successive vectors of a single feature stream. A series of experiments on phone recognition speaker-independent continuous-speech had been carried out using a subset of the large read-speech TIMIT phone corpus. Hidden Markov Model Toolkit (HTK) was also used throughout all carried experiments. Using such feature level combination, optimized mixture splitting and a bigram language model, a detailed analysis on this combination performance was discussed for Context-Independent (CI) and Context-Dependent (CD) Hidden Markov Models (HMM). Experimental results from our proposed procedure showed that phone error rate was successfully decreased by about 3%. At phonetic level group, an increase of 8% and of 10% was observed respectively for vowel and liquid group. These results proved clear phone enhancement regarding existing state of the art.

Robust speech recognition with articulatory features using dynamic Bayesian networks

Previous studies have proposed ways to estimate articulatory information from the acoustic speech signal and have shown that when used with standard cepstral features, they help to improve word recognition performance in noise for a connected digit recognition task. In this paper, I present results from a word recognition and a phone recognition experiments in noise that uses two sets of articulatory representation: continuous (tract variable trajectories) and discrete (articulatory gestures) along with standard mel cepstral features for acoustic modeling. The acoustic model is a dynamic Bayesian network (DBN) that treats the continuous articulatory information as observed and the discrete articulatory presentation as hidden random variables. Our results indicate that the use of articulatory information improved noise robustness for both the word recognition and phone recognition tasks substantially.

Non-symmetric time resolution for spectral feature trajectories

In a study presented at the fall 2010 meeting of the Acoustical Society of America (Zahorian etal., “Time/frequency resolution of acoustic features for automatic speech recognition”), we demonstrated that spectral/temporal evolution features which emphasize temporal aspects of acoustic features, with relatively low spectral resolution, are effective for phonetic recognition in continuous speech. These features are computed using discrete cosine transform coefficients for spectral information from 8 ms frames and discrete cosine series coefficients (DCSCs) for their temporal evolution, over overlapping intervals (blocks) longer than 200 ms. These features are presented as an alternative to mel-frequency cepstral coefficients, and their delta terms, for automatic speech recognition (ASR). In the present work, it is shown that these features are even more effective for ASR, using a non-symmetric time window which is tilted toward the beginning of each block when computing DCSCs. This non-symmetry can be implemented by combining two Gaussian windows with different standard deviations. This work also supports the hypothesis that the left context is somewhat more informative to phonetic identity than is the right context. Experimental results for automatic phone recognition are given for various conditions using the TIMIT and NTIMIT databases.

Spectral amplitude nonlinearities for improved noise robustness of spectral features for use in automatic speech recognition

Auditory models for outer periphery processing include a sigmoid shaped nonlinearity that is even more compressed than standard logarithmic scaling at very low and very high amplitudes. In some studies done at Carnegie Mellon University, it has been shown that this compressive nonlinearity is the most important aspect of the Seneff auditory model in terms of improving accuracy of automatic speech recognition in the presence of noise. However, in this previous work, the nonlinearity was trained for each frequency band of the Mel frequency cepstrum coefficients thus making it impractical to incorporate in automatic speech recognition systems. In the current study, a compressive nonlinearity is parametrically represented and constructed without training, to allow various degrees of steepness and “rounding” of corners for low and high amplitudes. Using this nonlinearity, experimental results for various noise conditions, and with mismatches in noise between training and test data, were obtained for phone recognition using the TIMIT and NTIMIT databases. The implications of the results are that a fixed compressive nonlinearity can be used to improve automatic speech recognition robustness with respect to mismatches between training and test data.

Discriminative Optimization of the Figure of Merit for Phonetic Spoken Term Detection

This paper proposes to improve spoken term detection (STD) accuracy by optimizing the figure of merit (FOM). In this paper, the index takes the form of a phonetic posterior-feature matrix. Accuracy is improved by formulating STD as a discriminative training problem and directly optimizing the FOM, through its use as an objective function to train a transformation of the index. The outcome of indexing is then a matrix of enhanced posterior-features that are directly tailored for the STD task. The technique is shown to improve the FOM by up to 13% on held-out data. Additional analysis explores the effect of the technique on phone recognition accuracy, examines the actual values of the learned transform, and demonstrates that using an extended training data set results in further improvement in the FOM.

Nonlinear enhancement of noisy speech, using continuous attractor dynamics formed in recurrent neural networks

Here, formation of continuous attractor dynamics in a nonlinear recurrent neural network is used to achieve a nonlinear speech denoising method, in order to implement robust phoneme recognition and information retrieval. Formation of attractor dynamics in recurrent neural network is first carried out by training the clean speech subspace as the continuous attractor. Then, it is used to recognize noisy speech with both stationary and nonstationary noise. In this work, the efficiency of a nonlinear feedforward network is compared to the same one with a recurrent connection in its hidden layer. The structure and training of this recurrent connection, is designed in such a way that the network learns to denoise the signal step by step, using properties of attractors it has formed, along with phone recognition. Using these connections, the recognition accuracy is improved 21% for the stationary signal and 14% for the nonstationary one with 0db SNR, in respect to a reference model which is a feedforward neural network.

A novel neural-based model for acoustic-articulatory inversion mapping

In this paper, a new bidirectional neural network for better acoustic-articulatory inversion mapping is proposed. The model is motivated by the parallel structure of human brain, processing information by having forward--reverse connections. In other words, there would be a feedback from articulatory system to the acoustic signals emitted from that organ. Inspired by this mechanism, a new bidirectional model is developed to map speech representations to articulatory features. Formation of attractor dynamics in such bidirectional model is first carried out by training the reference speaker subspace as the continuous attractor. Then, it is used to recognize the other speaker’s speech. In fact, the structure and training of this bidirectional model is designed in such a way that the network learns to denoise the signal step by step, using properties of attractors it has formed. In this work, the efficiency of a nonlinear feedforward network is compared to the same one with a bidirectional connection. The bidirectional model increases the accuracy up to approximately 3% (from 62.09 to 64.91%) in the phone recognition process.

Optimized discriminative transformations for speech features based on minimum classification error

Feature extraction is an important component of pattern classification and speech recognition. Extracted features should discriminate classes from each other while being robust to environmental conditions such as noise. For this purpose, several feature transformations are proposed which can be divided into two main categories: data-dependent transformation and classifier-dependent transformation. The drawback of data-dependent transformation is that its optimization criteria are different from the measure of classification error which can potentially degrade the classifier’s performance. In this paper, we propose a framework to optimize data-dependent feature transformations such as PCA (Principal Component Analysis), LDA (Linear Discriminant Analysis) and HLDA (Heteroscedastic LDA) using minimum classification error (MCE) as the main objective. The classifier itself is based on Hidden Markov Model (HMM). In our proposed HMM minimum classification error technique, the transformation matrices are modified to minimize the classification error for the mapped features, and the dimension of the feature vector is not changed. To evaluate the proposed methods, we conducted several experiments on the TIMIT phone recognition and the Aurora2 isolated word recognition tasks. The experimental results show that the proposed methods improve performance of PCA, LDA and HLDA transformation for mapping Mel-frequency cepstral coefficients (MFCC).

Time-Frequency Cepstral Features and Combining Discriminative Training for Phonotactic Language Recognition

The performance of the phonotactic system for language recognition depends on the quality of the phone recognizers. To improve the performance of the recognizers, this paper investigates the use of new acoustic features and discriminative training techniques for phone recognizers. The commonly used features are static ceptral coefficients appended with their first and second order deltas. This configuration may be not optimal for phone recognition in phonotactic language recognition systems. In this paper, a time-frequency cepstral (TFC) feature is proposed based on our previous work in acoustic language recognition systems. The feature is extracted as follows: first a temporal discrete cosine transform (DCT) is carried out on the cepstrum matrix, and then select the transformed elements in a specific area using the variance maximization criterion. Different parameters are tested to obtain the optimal configuration. Also, we adopt the feature minimum phone error (fMPE) method for discriminative training of phone models to obtain better phone recognition results for further improvement. The effectiveness of the two techniques is demonstrated on the NIST Language Recognition Evaluation (LRE) 2007database, including the 30 second, 10 second and 3 second closed-set test conditions.

Language Recognition Based on Acoustic Diversified Phone Recognizers and Phonotactic Feature Fusion

One typical phonotactic system for language recognition is parallel phone recognition followed by vector space modeling (PPRVSM). In this system, various phone recognizers are applied in parallel and fused at the score level. Each phone recognizer is trained for a known language, which is assumed to extract complementary information for effective fusion. But this method is limited by the large amount of training samples for which word or phone level transcription is required. Also, score fusion is not the optimal method as fusion at the feature or model level will retain more information than at the score level. This paper presents a new strategy to build and fuse parallel phone recognizers (PPR). This is achieved by training multiple acoustic diversified phone recognizers and fusing at the feature level. The phone recognizers are trained on the same speech data but using different acoustic features and model training techniques. For the acoustic features, Mel-frequency cepstral coefficients (MFCC) and perceptual linear prediction (PLP) are both employed. In addition, a new time-frequency cepstrum (TFC) feature is proposed to extract complementary acoustic information. For the model training, we examine the use of the maximum likelihood and feature minimum phone error methods to train complementary acoustic models. In this study, we fuse phonotactic features of the acoustic diversified phone recognizers using a simple linear fusion method to build the PPRVSM system. A novel logistic regression optimized weighting (LROW) approach is introduced for fusion factor optimization. The experimental results show that fusion at the feature level is more effective than at the score level. And the proposed system is competitive with the traditional PPRVSM. Finally, the two systems are combined for further improvement. The best performing system reported in this paper achieves an equal error rate (EER) of 1.24%, 4.98% and 14.96% on the NIST 2007 LRE 30-second, 10-second and 3-second evaluation databases, respectively, for the closed-set test condition.

Time/frequency resolution of acoustic features for automatic speech recognition.

The underlying assumption for spectral/temporal features for use in automatic speech recognition is that the frequency resolution should be emphasized in relation to temporal resolution. Accordingly, Mel frequency cepstral coefficients are typically computed using an approximately 25‐ms frame length with a 10‐ms frame spacing, and using 3–5 frames to represent temporal derivative information. In phone recognition experiments based on the TIMIT database using discrete cosine transform coefficients for spectral information and discrete cosine series coefficients for their temporal evolution, substantially higher phone accuracies were obtained with much shorter frame lengths (8 ms), much shorter frame spacings (2 ms), and much longer time intervals for capturing spectral tracks (on the order of 500 ms). Experimental results with various conditions are given for phone recognition using the TIMIT database. The implications of these results are that spectral/temporal evolution features, emphasizing the temporal aspects, are of great importance for automatic speech recognition.

Enhanced Phone Posteriors for Improving Speech Recognition Systems

Using phone posterior probabilities has been increasingly explored for improving automatic speech recognition (ASR) systems. In this paper, we propose two approaches for hierarchically enhancing these phone posteriors, by integrating long acoustic context, as well as phonetic and lexical knowledge. In the first approach, phone posteriors estimated with a multilayer perceptron (MLP), are used as emission probabilities in hidden Markov model (HMM) forward-backward recursions. This yields new enhanced posterior estimates integrating HMM topological constraints (encoding specific phonetic and lexical knowledge), and context. In the second approach, temporal contexts of the regular MLP posteriors are postprocessed by a secondary MLP, in order to learn inter- and intra-dependencies between the phone posteriors. These dependencies are phonetic knowledge. The learned knowledge is integrated in the posterior estimation during the inference (forward pass) of the second MLP, resulting in enhanced phone posteriors. We investigate the use of the enhanced posteriors in hybrid HMM/artificial neural network (ANN) and Tandem configurations. We propose using the enhanced posteriors as replacement, or as complementary evidences to the regular MLP posteriors. The proposed methods have been tested on different small and large vocabulary databases, always resulting in consistent improvements in frame, phone, and word recognition rates.