Digit Speech Recognition Research Articles

A strategic approach to recognize the speech of the children with hearing impairment: different sets of features and models

The automatic speech recognition system is developed and tested for recognizing the speeches of a normal person in various languages. This paper mainly emphasizes the need for the development of a more challenging speaker independent speech recognition system for hearing impaired to recognize the speeches uttered by any Hearing Impaired (HI) speaker. In this work, Gamma tone energy features with filters spaced an equivalent rectangular bandwidth (ERB), MEL & BARK scale, and MFPLPC features are used at the front end and vector quantization (VQ) & multivariate hidden Markov models (MHMM) at the back end for recognizing the speeches uttered by any hearing impaired speaker. Performance of the system is compared for the three modeling techniques VQ, FCM (Fuzzy C means) clustering and MHMM for the recognition of isolated digits and simple continuous sentences in Tamil. Recognition accuracy (RA) is 81.5% with speeches of eight speakers considered for training and speeches of the remaining two speakers considered for testing for speaker independent isolated digit recognition system. Accuracy is found to be 91% and 87.5% for considering 90% of the data for training and 10% for testing for speaker independent isolated digit and continuous speech recognition systems respectively. Accuracy can be further enhanced by having an extensive database for creating models/templates. Receiver operating characteristics (ROC) drawn between True Positive Rate and False Positive Rate is used to assess the performance of the system for HI. This system can be utilized to understand the speech uttered by any hearing impaired speaker and the system facilitates the provision of necessary assistance to them. It ultimately improves the social status of the hearing impaired people and their confidence level will be enhanced.

Auditory Cognitive Training for Pediatric Cochlear Implant Recipients.

Understanding speech in noise is the biggest challenge faced by individuals with cochlear implants (CIs). Improving speech-in-noise perception for pediatric CI recipients continues to remain a high priority for all stakeholders. This study was designed to investigate the efficacy of working memory training for improving speech-in-noise recognition for children with CIs. Fourteen children with CIs (aged 6 to 15 years) received adaptive, home-based training on forward digit span task for 5 weeks, while 13 children with CIs participated in backward digit span training. Seventeen age-matched children with CIs in the control group received a placebo. Outcome measures included forward and backward digit span and speech recognition threshold for digit triplets in noise at pretraining, post-training, and 5-week follow-up. Performance measures from 26 age-matched children with normal hearing were also obtained only at the baseline session to generate normative standards for comparison. Digit span scores were significantly higher at post-training and follow-up compared with pretraining for both forward- and backward-trained groups. Both trained groups showed similar training-induced shifts. The control group showed no such improvement. There was no effect of training on the speech recognition threshold. Children with CIs showed significantly lower digit span scores and a higher speech recognition threshold relative to children with normal hearing. Training improves working memory capacity as measured by digit spans for children with CIs. Training-induced improvements are stable for at least 5 weeks. Learning effects demonstrate near-transfer, from forward to backward digit span and vice versa, but failed to show far-transfer to speech-in-noise recognition. Current evidence is not supportive of cognitive training for improving speech-in-noise performance for children with CIs.

An experimental framework for Arabic digits speech recognition in noisy environments

In this paper we present an experimental framework for Arabic isolated digits speech recognition named ARADIGITS-2. This framework provides a performance evaluation of Modern Standard Arabic devoted to a Distributed Speech Recognition system, under noisy environments at various Signal-to-Noise Ratio (SNR) levels. The data preparation and the evaluation scripts are designed by deploying a similar methodology to that followed in AURORA-2 database. The original speech data contains a total of 2704 clean utterances, spoken by 112 (56 male and 56 female) Algerian native speakers, down-sampled at 8 kHz. The feature vectors, which consist of a set of Mel Frequency Cepstral Coefficients and log energy, are extracted from speech samples using ETSI Advanced Front-End (ETSI-AFE) standard; whereas, the Hidden Markov Models (HMMs) Toolkit is used for building the speech recognition engine. The recognition task is conducted in speaker-independent mode by considering both word and syllable as acoustic units. Therefore, an optimal fitting of HMM parameters, as well as the temporal derivatives window, is carried out through a series of experiments performed on the two training modes: clean and multi-condition. Better results are obtained by exploiting the polysyllabic nature of Arabic digits. These results show the effectiveness of syllable-like unit in building Arabic digits recognition system, which exceeds word-like unit by an overall Word Accuracy Rate of 0.44 and 0.58% for clean and multi-condition training modes, respectively.

Introduction of digital speech recognition in a specialised outpatient department: a case study

BackgroundSpeech recognition software might increase productivity in clinical documentation. However, low user satisfaction with speech recognition software has been observed. In this case study, an approach for implementing a speech recognition software package at a university-based outpatient department is presented.MethodsMethods to create a specific dictionary for the context “sports medicine” and a shared vocabulary learning function are demonstrated. The approach is evaluated for user satisfaction (using a questionnaire before and 10 weeks after software implementation) and its impact on the time until the final medical document was saved into the system.ResultsAs a result of implementing speech recognition software, the user satisfaction was not remarkably impaired. The median time until the final medical document was saved was reduced from 8 to 4 days.ConclusionIn summary, this case study illustrates how speech recognition can be implemented successfully when the user experience is emphasised.Electronic supplementary materialThe online version of this article (doi:10.1186/s12911-016-0374-4) contains supplementary material, which is available to authorized users.

High-order hidden Markov model for piecewise linear processes and applications to speech recognition.

The hidden Markov models have been widely applied to systems with sequential data. However, the conditional independence of the state outputs will limit the output of a hidden Markov model to be a piecewise constant random sequence, which is not a good approximation for many real processes. In this paper, a high-order hidden Markov model for piecewise linear processes is proposed to better approximate the behavior of a real process. A parameter estimation method based on the expectation-maximization algorithm was derived for the proposed model. Experiments on speech recognition of noisy Mandarin digits were conducted to examine the effectiveness of the proposed method. Experimental results show that the proposed method can reduce the recognition error rate compared to a baseline hidden Markov model.

The development and evaluation of the Finnish digit triplet test

Objectives: The aim of the study was to develop a reliable and easily accessible screening test for primary detection of hearing impairment.Methods: Digits 0–9 were used to form quasirandom digit triplets. First, digit specific intelligibility functions and speech recognition thresholds (SRTs) were determined. To homogenize the test material digits with steep intelligibility function slopes were chosen and level correction up to ±2 dB were applied to the digits as needed. Evaluation measurements were performed to check for systematic differences in intelligibility between the test lists and to obtain normative reference function for normal-hearing listeners.Results: The mean SRT and the final slope of the test lists were −10.8 ± 0.1 dB signal-to-noise ratio (SNR) and 21.7 ± 1.8%/dB, respectively (measurements at constant level; inter-list variability). The mean SRT and slope of the test subjects were −10.8 ± 0.5 dB SNR and 23.4 ± 5.2%/dB (measurements at constant level; inter-subject variability). The mean SRT for normal-hearing young adults for a single adaptive measurement is −9.8 ± 0.9 dB SNR.Conclusion: The Finnish digit triplet test is the first self-screening hearing test in the Finnish language. It was developed according to current standards, and it provides reliable and internationally comparable speech intelligibility measurements.

Quality management in musculoskeletal imaging: form, content, and diagnosis of knee MRI reports and effectiveness of three different quality improvement measures.

The purpose of this article is to evaluate the quality of reports of knee MRI examinations in form, content, and diagnosis and to assess the effect of three different quality improvement measures. Reports of 500 knee MRI examinations (first round, or baseline) were reviewed retrospectively by five musculoskeletal radiologists. Fifteen different criteria were assessed for formal and content-related quality of reports. Diagnostic discrepancies were categorized using a 5-point scale, as follows: I, no deviation; II, undetected finding, clinically irrelevant; III, wrong interpretation of finding, clinically irrelevant; IV, undetected finding, clinically relevant; and V, wrong interpretation of finding, clinically relevant. Then three different quality improvement measures were applied prospectively to a total of 510 consecutive reports: a quiet work environment, double reading, and the use of a structured report template. These 510 knee MRI reports (second round) were evaluated using the same criteria. A statistically significant improvement in 13 of 15 criteria was found in the second-round reports: orthographic errors improved from 32.4% to 22.0% (p < 0.001) of reports and digital speech recognition errors improved from 8.4% to 7.6% (p = 0.660). Missing anatomic structures decreased from 6.3% to 0.4%. Diagnostic discrepancies were less frequently found (12.9% vs 20.8%; p = 0.001), along with the following changes in categorization (first-round results are in parentheses): I, 87.1% (79.2%); II, 9.2% (16.8%); III, 2.3% (1.0%); IV, 1.0% (2.6%); and V, 0.4% (0.4%). Quality improvement was found in all three measure groups, without statistical significance among the groups, except for the orthographic errors (p < 0.001), which were most common in the template group. All three quality improvement measures significantly improved the quality of the knee MRI reports, but no measure was clearly superior to the others.

Realization of Hidden Markov Model for English Digit Recognition

objective of the work described here is to compare the isolated English language digit speech recognition using Hidden Markov Model for speaker independent system. Two different datasets were collected of audio recordings for the said comparison of isolated digits of English language. Speakers here read numeric digits 0 to 9 i.e. ZERO to NINE. One corpus is self recorded signals and other is standard CUAVE dataset (36 speakers, each uttered 10 words). The training and testing samples are separated for speaker dependent and speaker independent systems. The system has been implemented using the HMM toolkit i.e. HTK by training HMMs of the words making the vocabulary on the training data. Different HMMs for individual digits have been initialized and trained to have well modeled structure. The trained system was tested on training data as well as test data and results shown that most of the speech samples were correctly recognized. The system was tested for speaker independent and dependent way, to check the changes in the recognition rate. Further this can be used by developers and researchers interested in speech recognition for English language not only for isolated digits but also for other words of English language. If clean database is available, further this can be generalized to recognize words of any language. Continuous speech can also be recognized using study of this system.

Digit Speech Recognition Threshold (SRT) in Children with Normal Hearing Ages 5–8 years

The purpose of this study was to compare digit stimuli to popular spondees when measuring speech recognition threshold (SRT) in children with normal hearing and to determine the influence of increment size using a descending method (ASHA, 1988). SRT was measured with paired digit stimuli (D-SRT) and pediatric word stimuli (W-SRT) using 2 step sizes (5 dB, 2 dB) in 30 typically developing children with normal hearing ages 5;0 (years;months) to 8;11. Pure-tone and SRT tests were administered in both ears. The D-SRT and W-SRT were compared to pure-tone average (PTA) thresholds per ear. Analysis of variance revealed a significant interaction of test and step size: PTA was significantly lower when measured with 2-dB than with 5-dB increment; however, step size yielded no differences in D-SRT and W-SRT. Right ear thresholds across tests were significantly lower than left ear. Pearson correlations were significant for all tests except two; D-SRT was highly correlated to W-SRT (r = .49-.72) in both ears. Regression analyses for both ears and step sizes revealed that prediction of the hearing threshold for speech from PTA was equally accurate for SRT measured with digit pairs or popular pediatric spondee stimuli. Digit pairs are an appropriate alternative stimulus for SRT measurement in children with normal hearing. Future diagnostic audiology implications, including application for other pediatric populations, are discussed.

Scalable Semi-Supervised Classification via Neumann Series

Traditional graph-based semi-supervised learning (GBSSL) algorithms usually scale badly due to the expensive computational burden. The main bottleneck is that they need to compute the inversion of a huge matrix. In order to alleviate this problem, this paper proposes Neumann series approximation (NSA) to explicitly approximate the inversion process required by conventional GBSSL methodologies, which makes them computationally tractable for relatively large datasets. It is proved that the deviation between the approximation and direct inversion is bounded. Using real-world datasets related to handwritten digit recognition, speech recognition and text classification, the experimental results reveal that NSA accelerates the speed significantly without decreasing too much precision. We also empirically show that NSA outperforms other scalable approaches such as Nystrom method, Takahashi equation, Lanczos process based SVD and AnchorGraph regularization, in terms of both efficiency and accuracy.

Connected digit speech recognition system for Malayalam language

A connected digit speech recognition is important in many applications such as automated banking system, catalogue-dialing, automatic data entry, automated banking system, etc. This paper presents an optimum speaker-independent connected digit recognizer for Malayalam language. The system employs Perceptual Linear Predictive (PLP) cepstral coefficient for speech parameterization and continuous density Hidden Markov Model (HMM) in the recognition process. Viterbi algorithm is used for decoding. The training data base has the utterance of 21 speakers from the age group of 20 to 40 years and the sound is recorded in the normal office environment where each speaker is asked to read 20 set of continuous digits. The system obtained an accuracy of 99.5 % with the unseen data.

Connected Mandarin Digit Speech Recognition Using Two-Layer Acoustic Universal Structure

Because of the single-syllable of Chinese words and the confusing nature of Chinese pronunciation, connected mandarin digit speech recognition (CMDSR) is a challenging task in the field of speech recognition. This paper applied a novel acoustic representation of speech, called the acoustic universal structure (AUS) where the non-linguistic variations such as vocal tract length, lines and noises are well removed. A two-layer matching strategy based on the AUS models of speech, including the digit and string AUS models, is proposed for connected mandarin digit speech recognition. The speech recognition system for connected mandarin digits is described in detail, and the experimental results show that the proposed method can obtain the higher recognition rate.

Review of distinctive phonetic features and the Arabic share in related modern research

Most research in the field of digital speech technology has traditionally been conducted in only a few languages, such as English, French, Spanish, or Chinese. Numerous studies using distinctive phonetic features (DPFs) with different techniques and algorithms have been carried out during the last 3 decades, mainly in English, Japanese, and other languages of industrialized countries. DPF elements are based on a technique used by linguists and digital speech and language experts to distinguish between different phones by considering the lowest level of actual features during phonation. These studies have investigated the best performances, outcomes, and theories, especially those regarding digital speech recognition. The aim of this paper is to present the background of DPF theories and the usefulness thereof for digital speech and language processing. In addition, we highlight the background of Arabic language phonology compared to 2 well-known languages to enhance the current knowledge about this narrow language discipline. Finally, this work reviews the research dealing with DPF strategies for digital speech and language processing using computing and engineering techniques and theories. Based on the literature search conducted for this paper, we conclude that although the Arabic language is a very important and old Semitic language, hitherto it has suffered from a lack of modern research resources and theories on DPF elements.

Neighboring Digits Pattern Training Method in Quickly-spoken Connected Mandarin Digits Speech Recognition

Deletion errors are most usually occurred in connected Mandarin digit string speech recognition when speaking rate is fast, and are the main reasons leading to the increasing of the recognition error rate and the decline of the recognition accuracy. In this paper, a new training method named neighboring digits pattern is given based on sufficient statistics of recognition errors of the traditional system in order to eliminate most of deletion errors which seriously affect the system recognition rate. The training process is presented and the performance evaluation is given. The result analysis demonstrates that the new method can reduce the deletion errors effectively and improve the system recognition rate from 96.4% to 98.3%.

Combining Speech Enhancement and Cepstral Mean Normalization for LPC Cepstral Coefficients

A mismatch between the training and testing in noisy circumstance often causes a drastic decrease in the performance of speech recognition system. The robust feature coefficients might suppress this sensitivity of mismatch during the recognition stage. In this paper, we investigate the noise robustness of LPC Cepstral Coefficients (LPCC) by using speech enhancement with feature post-processing. At front-end, speech enhancement in the wavelet domain is used to remove noise components from noisy signals. This enhanced processing adopts the combination of discrete wavelet transform (DWT), wavelet packet decomposition (WPD), multi-thresholds processing etc to obtain the estimated speech. The feature post-processing employs cepstral mean normalization (CMN) to compensate the signal distortion and residual noise of enhanced signals in the cepstral domain. The performance of digit speech recognition systems is evaluated under noisy environments based on NOISEX-92 database. The experimental results show that the presented method exhibits performance improvements in the adverse noise environment compared with the previous features.

Bayesian Separation With Sparsity Promotion in Perceptual Wavelet Domain for Speech Enhancement and Hybrid Speech Recognition

Speech recognition accuracy can be improved by the removal of noise. However, errors in the estimated signal components can also obscure the recognition. This paper presents a framework of wavelet-based techniques to harness the automatic speech recognition performance in the presence of background noise. The proposed robust speech recognition system is realized by implementing speech enhancement preprocessing, feature extraction, and a hybrid speech recognizer in the time-frequency space. A perceptual wavelet filterbank using a fixed base to imitate the human perceptual modus of speech is developed to capture the most discriminative information in the time-frequency plane. To minimize the mismatch between the training and testing conditions of the classifier, a Bayesian scheme is applied in a wavelet domain to separate the speech and noise components in the proposed iterative speech enhancement algorithm. The nonphonetic information is discarded while the more critical speech features are extracted and represented by the wavelet coefficients. The denoised wavelet features are fed to the hybrid classifier founded on a hidden Markov model (HMM). The intrinsic limitation of the HMM is overcome by augmenting it with a wavelet support vector machine. This hybrid and hierarchical design paradigm improves the recognition performance by combining the advantages of different methods into an integral system. The continuous digit speech recognition experiments conducted with the proposed framework show promising results. It significantly improves the recognition performance at a low signal-to-noise ratio (SNR) without causing a poorer performance at a high SNR.

Formant analysis in dysphonic patients and automatic Arabic digit speech recognition

Background and objectiveThere has been a growing interest in objective assessment of speech in dysphonic patients for the classification of the type and severity of voice pathologies using automatic speech recognition (ASR). The aim of this work was to study the accuracy of the conventional ASR system (with Mel frequency cepstral coefficients (MFCCs) based front end and hidden Markov model (HMM) based back end) in recognizing the speech characteristics of people with pathological voice.Materials and methodsThe speech samples of 62 dysphonic patients with six different types of voice disorders and 50 normal subjects were analyzed. The Arabic spoken digits were taken as an input. The distribution of the first four formants of the vowel /a/ was extracted to examine deviation of the formants from normal.ResultsThere was 100% recognition accuracy obtained for Arabic digits spoken by normal speakers. However, there was a significant loss of accuracy in the classifications while spoken by voice disordered subjects. Moreover, no significant improvement in ASR performance was achieved after assessing a subset of the individuals with disordered voices who underwent treatment.ConclusionThe results of this study revealed that the current ASR technique is not a reliable tool in recognizing the speech of dysphonic patients.

A Multi-Space Distribution (MSD) and two-stream tone modeling approach to Mandarin speech recognition

Tone plays an important role in recognizing spoken tonal languages like Chinese. However, the discontinuity of F0 between voiced and unvoiced transition has traditionally been a hurdle in creating a succinct statistical tone model for automatic speech recognition and synthesis. Various heuristic approaches have been proposed before to get around the problem but with limited success. The Multi-Space Distribution (MSD) proposed by Tokuda et al. which models the two probability spaces, discrete for unvoiced region and continuous for voiced F0 contour, in a linearly weighted mixture, has been successfully applied to Hidden Markov Model (HMM)-based text-to-speech synthesis. We extend MSD to Chinese Mandarin tone modeling for speech recognition. The tone features and spectral features are further separated into two streams and corresponding stream-dependent models are trained. Finally two separated decision trees are constructed by clustering corresponding stream-dependent HMMs. The MSD and two-stream modeling approach is evaluated on large vocabulary, continuously read and spontaneous speech Mandarin databases and its robustness is further investigated in a noisy, continuous Mandarin digit database with eight types of noises at five different SNRs. Experimental results show that our MSD and two-stream based tone modeling approach can significantly improve the recognition performance over a toneless baseline system. The relative tonal syllable error rate (TSER) reductions are 21.0%, 8.4% and 17.4% for large vocabulary read and spontaneous and noisy digit speech recognition tasks, respectively. Comparing with the conventional system where F0 contours are interpolated in unvoiced segments, our approach improves the recognition performance by 9.8%, 7.4% and 13.3% in relative TSER reductions in the corresponding speech recognition tasks, respectively.

Compact hardware liquid state machines on FPGA for real-time speech recognition

Hardware implementations of Spiking Neural Networks are numerous because they are well suited for implementation in digital and analog hardware, and outperform classic neural networks. This work presents an application driven digital hardware exploration where we implement real-time, isolated digit speech recognition using a Liquid State Machine. The Liquid State Machine is a recurrent neural network of spiking neurons where only the output layer is trained. First we test two existing hardware architectures which we improve and extend, but that appears to be too fast and thus area consuming for this application. Next, we present a scalable, serialized architecture that allows a very compact implementation of spiking neural networks that is still fast enough for real-time processing. All architectures support leaky integrate-and-fire membranes with exponential synaptic models. This work shows that there is actually a large hardware design space of Spiking Neural Network hardware that can be explored. Existing architectures have only spanned part of it.

Robust Signal Subspace Speech Classifier

A speech model inspired by the signal subspace approach was recently proposed as a speech classifier with modest results. The method entails, in general, the assemblage of a set of subspace trajectories that consist of the right singular vectors of measurement matrices of the signal under consideration. Given an unknown signal, a simple distortion measure then applies in the classification procedure to pick the best matched class prototype. This letter examines the issue of robustness in the subspace classification scheme. Borrowing an important result on noisy measurement matrices, this letter formally establishes the notion of robustness in subspace classification and proceeds to propose a class of robust distortion measures for signal subspace models. Simulation results of subspace classifiers implementing the new distortion measures in an isolated digit speech recognition problem reveal no degradation in recognition accuracy, even under low SNR conditions.