Hindi Phoneme Research Articles

Perception of Voicing and Aspiration in Hindi, American English, and Tamil Listeners in Quiet and in Background Noise.

There is a dearth of literature in determining whether language groups for whom aspiration and/or voicing is phonologically contrastive show better perception relative to those who do not use these features contrastively and whether the cue type modulates perception in noise. This study addresses perception of laryngeal cues (voicing and aspiration) by Hindi, English, and Tamil listeners, in quiet and in noise. Sixteen participants between 20 and 45 years of age were included in each of the three language groups. The stimuli were bilabial stops that contrasted phonetically in voicing and aspiration, voicing-lead [ba], short-lag [pa], and long-lag aspirated [pha], with one set corresponding to the Hindi phonemes /ba/, /pa/, and /pha/ and the second set to the English phonemes /ba/ and /pa/ (which are phonetically [pa] and [pha], respectively). Tamil includes only the short-lag [pa] as a bilabial stop consonant. The stimuli were presented at 70 dB SPL, in quiet and in speech-shaped noise at a signal-to-noise ratio of 0. Participants performed two speech identification tasks and a speech discrimination task. Patterns of perceptual assimilation related to the first language were observed in all three language groups, and accuracy was generally higher in quiet than in noise. Hindi participants identified the English /pa/ as Hindi /pha/ and English /ba/ as Hindi /pa/. The American English participants identified Hindi /pha/ as English /pa/ and both the Hindi /pa/ and Hindi /ba/ as English /ba/. In contrast, Tamil listeners generally perceived both Hindi and English bilabial stops as one category, regardless of voicing and aspiration. English and Hindi participants generally showed higher accuracy for native language stimuli. Patterns of assimilation in quiet and noise differed across language groups for each stimulus type. The aspirated stimuli were most likely to be misperceived in noise by all groups (often as /ha/). The results serve as evidence that listeners accurately access native language speech cues, even in noise. The results contribute toward a better understanding of cross-linguistic speech processing in noise.

Confusion analysis in phoneme based speech recognition in Hindi

Phoneme recognition is an essential step in the development of a speech recognition system (SRS), as phonemes are fundamental building blocks in a spoken language. This research work aimed to present phoneme recognition with systematic confusion analysis for the Hindi language. The accuracy of phoneme recognition is the foundation for developing an efficient SRS. Therefore, the systematic confusion analysis for phoneme recognition is essential to improve speech recognition performance. Experiments conducted on Continuous Hindi speech corpus for phoneme recognition with speaker-dependent mode using Hidden Markov Model (HMM) based tool kit HTK. Feature extraction technique Perceptual Linear Predictive Coefficient (PLP) was used with five states Monophones HMM model. Tests were performed for exploring the recognition of Hindi vowels and consonants. Confusion matrices were presented for both vowels and consonants with analysis and possible solutions. During systematic analysis, the vowels were divided into front, middle, and back vowels while consonants were categorized based on place of articulation and manner of articulation. Research findings show that some Hindi phonemes have significant effects on speech recognition. The investigations also reveal that some Hindi phonemes are mostly confused, and some phonemes have more deletions and insertions. The research further demonstrates that the words made of less number of phonemes show more insertion errors. It was also found that most of the Hindi sentences end with some specific words. These particular words can be used to reduce the search place in language modeling for improving speech recognition. The research findings can be utilized to enhance the performance of the speech recognition system by selecting suitable feature extraction techniques and classification techniques for phonemes. The outcome of the research can also be used to develop improved pronunciation dictionaries and designing the text for developing phonetically balanced speech corpus for improvement in speech recognition. Experimental results show an average corrected recognition score of 70% for vowel class and consonant categories, the maximum average corrected recognition score of 94% was obtained with palatal sounds, and the lowest average corrected recognition score of 54% was achieved with liquid sounds. The comparative analysis of the presented work was made to similar existing works.

The perceptual assimilation and discrimination of 20 Hindi consonants by native speakers of English

Hindi consonant contrasts are known to pose difficulties for native English speakers (e.g., Polka, 1991; Werker and Tees, 2002; and Cibelli, 2015). While a limited set of Hindi consonants have received a great deal of attention (i.e., the coronal stops), we do not yet know how learners perceive other segments from Hindi’s relatively large consonant inventory. To address this gap, we conducted a perceptual assimilation study modeled after Faris et al. (2018) to investigate the perceptual assimilation of twenty Hindi consonants (tʃ, tʃʰ, dʒ, dʒʰ, t, tʰ, d, dʰ, ʈ, ʈʰ, ɖ, ɖʰ, ʃ, ʂ, s, z, l, r, ɽ, ɽʰ) by native English speakers with no prior Hindi language learning experience. We later examined the discrimination of 46 pairs from the same set of phones in a new group of participants from the same population using an AX discrimination task. Participants exhibited patterns of perceptual assimilation and discrimination of Hindi phonemes that suggest that native English speakers will experience difficulty across the Hindi consonant inventory. We consider the findings in the context of the Perceptual Assimilation Model (Best, 1994 and Best and Tyler, 2007), providing a fuller account of the difficulty posed by the Hindi consonant inventory for second-language learners.Hindi consonant contrasts are known to pose difficulties for native English speakers (e.g., Polka, 1991; Werker and Tees, 2002; and Cibelli, 2015). While a limited set of Hindi consonants have received a great deal of attention (i.e., the coronal stops), we do not yet know how learners perceive other segments from Hindi’s relatively large consonant inventory. To address this gap, we conducted a perceptual assimilation study modeled after Faris et al. (2018) to investigate the perceptual assimilation of twenty Hindi consonants (tʃ, tʃʰ, dʒ, dʒʰ, t, tʰ, d, dʰ, ʈ, ʈʰ, ɖ, ɖʰ, ʃ, ʂ, s, z, l, r, ɽ, ɽʰ) by native English speakers with no prior Hindi language learning experience. We later examined the discrimination of 46 pairs from the same set of phones in a new group of participants from the same population using an AX discrimination task. Participants exhibited patterns of perceptual assimilation and discrimination of Hindi phonemes that suggest that native English speakers will experience difficulty acro...

Wavelet-Based Power Normalized Spectrum for Hindi Phoneme Classification

This paper presents wavelet-based power normalized spectrum for computing robust cepstral features named WP-PNCC features. The proposed technique computes wavelet packet-based short-time spectrum of speech signal. A nonlinear function is defined as relating power spectrum of clean speech to the power spectrum of speech corrupted with noise. The constants of function are computed from longer-duration speech spectrum, and the short-time spectrum for each frame is weighted with the power function. The weighted speech spectrum is processed with logarithmic and discrete cosine transform operation to compute cepstral coefficients. The cepstral coefficients thus obtained are processed with quantile-based cepstral dynamics normalization technique. The proposed features are examined with hidden Markov model classifier on TIFR database for Hindi phoneme classification task and on TIMIT database for English phoneme classification task along with mel-frequency cepstral coefficients, power normalized cepstral coefficients and 24-band wavelet-based features in clean and noisy environments. Different noises from NOISEX-92 database are used for preparing noisy database with SNR ranging from 20 dB to 0 dB. The results show enhanced performance of proposed features in all the considered cases. The simulations are performed on MATLAB 2015b. The performance of proposed features is also evaluated on hidden Markov model toolkit-based speech recognition system. The comparative results confirm the robustness of proposed features with sufficient improvement over other features examined in this paper.

Nvestigations of the Approximate Percentage of Noise Required to Perceive Hindi Phonemes using HNM

nvestigations of the Approximate Percentage of Noise Required to Perceive Hindi Phonemes using HNM

Influence of Hindi Cartoons on the Language Development of Pakistani Urdu/English Schoolchildren

Not surprisingly, children expand their linguistic repertoire with the maturation of their cognition and their daily activities such as watching cartoons influence their language development. Many if not all parents fear that the Hindi cartoons are contaminating the linguistic repertoire of their children. Eventually, Pakistan Media Electronic Regularity Authority has banned Hindi dubbed cartoons. However, it is quite paradoxical that they always wish to improve the English language skills of their children but they have negative grounding towards the Hindi language. With this background in minds, the present study was carried out to examine the influence of some popular Hindi cartoons (such as Doraemon, Motu Patlu, and Chota Bheem) on the phonological development of children from middle and upper-middle class schools located in Lahore. By taking the Labovian apparent hypothesis as a theoretical lens, a sample of 84 Urdu/English bilinguals of different age groups was selected to investigate a prevalence of most commonly occurring phonological variants (such as /kʰ/ for /x/, /g/ for /ɣ/, / d ʒ/ for /z/, and /f/ for /pʰ/) in their speech. The study revealed that the influence of the Hindi phonology remained temporary on the language development of the participants. They were open to the usage of Hindi phonemes unless they developed alternatives phonological system in their first language (Urdu) and second and the most prestigious language (English). The participants who were around 10 years of age were found relatively less prone to using the Hindi phones. We may not rule out completely whether or not the above-mentioned phonological variants were part of their linguistic repertoire when they reached to puberty. It was observed, on the other hand, that some low-level phonological variations recycled invariably in the speech of the participants. Hence, the study suggests investigating these variations beyond this age bracket. The study takes a diametrically opposite stance and suggests an exposure to multilingual vocabulary as a correlate of cognitive development during the early childhood education.

Classification of myoelectric signal for sub-vocal Hindi phoneme speech recognition

Classification of myoelectric signal for sub-vocal Hindi phoneme speech recognition

Does it have to be correct?: The effect of uninformative feedback on non-native phone discrimination

Learning the phonetic inventory of a non-native language requires perceptual adjustment to non-native phones that sometimes belong to a single category in the learner’s native language. For example, English native speakers often struggle to learn the distinction between the Hindi phonemes [ʈ] and [t̪] that are both categorized as [t] in American English (AE). Olmstead and Viswanathan (2017) showed that AE listener’s discrimination of these non-native phones could be improved using short exposure to naturally produced Indian English (IE) words that contained the target contrast. In the current study, we examine how feedback affects this lexical retuning effect. Specifically, we set up feedback schedules that either reinforce the consistent mapping of these consonants onto the AE speaker’s existing [t] and [θ], or that reinforce an inconsistent mapping. If the consistency of this mapping in IE is paramount to improving phonetic discrimination, then reinforcing it should strengthen the effect and providing a variable mapping should weaken it. Results suggest that the feedback given to participants did not change the effect on discrimination—discrimination improved whether or not the feedback was consistent. Implications of these findings are discussed.

Neurophysiological aspects of brainstem processing of speech stimuli in audiometric-normal geriatric population.

Poor auditory speech perception in geriatrics is attributable to neural de-synchronisation due to structural and degenerative changes of ageing auditory pathways. The speech-evoked auditory brainstem response may be useful for detecting alterations that cause loss of speech discrimination. Therefore, this study aimed to compare the speech-evoked auditory brainstem response in adult and geriatric populations with normal hearing. The auditory brainstem responses to click sounds and to a 40 ms speech sound (the Hindi phoneme |da|) were compared in 25 young adults and 25 geriatric people with normal hearing. The latencies and amplitudes of transient peaks representing neural responses to the onset, offset and sustained portions of the speech stimulus in quiet and noisy conditions were recorded. The older group had significantly smaller amplitudes and longer latencies for the onset and offset responses to |da| in noisy conditions. Stimulus-to-response times were longer and the spectral amplitude of the sustained portion of the stimulus was reduced. The overall stimulus level caused significant shifts in latency across the entire speech-evoked auditory brainstem response in the older group. The reduction in neural speech processing in older adults suggests diminished subcortical responsiveness to acoustically dynamic spectral cues. However, further investigations are needed to encode temporal cues at the brainstem level and determine their relationship to speech perception for developing a routine tool for clinical decision-making.

Admissible wavelet packet sub‐band based harmonic energy features using ANOVA fusion techniques for Hindi phoneme recognition

Nowadays, wavelet packet (WP) based features have been used extensively to maximise the performance of automatic speech recognition in the complex auditory environment. However, wavelet features are less sufficient to represent the voiced speech. Recent researches on WP technique seek for complementary voicing information to overcome this problem. However, considering additional voicing features results in longer dimension and somehow affected the performance for unvoiced speech. This study presents a new analysis of variance technique to incorporate voicing information on WP sub-band based features without affecting its performance and dimension. It has been noticed that most of the voiced energy lies below 2 kHz. Thus, the proposed technique emphasises the lower sub-bands for additional voicing information. Harmonic energy features are combined with recently introduced auditory motivated equivalent rectangular bandwidth like 24-band WP cepstral features to enhance the performance of voiced phoneme recogniser. Primarily, a standard phonetically balanced Hindi database is used to analyse the performance of the proposed technique across a wide range of signal-to-noise ratios. Proposed features show a promising result in phoneme recognition experiment without affecting the feature dimension and performance.

Sub-vocal speech pattern recognition of Hindi alphabet with surface electromyography signal

Summary Recently electromyography (EMG) based speech signals have been used as pattern recognition of phoneme, vocal frequency estimation, browser interface, and classification of speech related problem identification. Attempts have been made to use EMG signal for sub-vocal speech pattern recognition of Hindi phonemes and Hindi words. That provides the command sub-vocally to control the devices. Sub-vocal EMG data were collected from more than 10 healthy subjects aged between 25 and 30 years. EMG-based sub-vocal database are acquired from four channel BIOPAC MP-30 acquisition system. Four pairs of Ag-AgCl electrodes placed in the participant neck area of skin. AR coefficients and Cepstral coefficients were computed as features of EMG-based sub-vocal signal. Furthermore, these features are classified by HMM classifier. H2M MATLAB toolbox was used to develop HMM classifier for classification of phonemes. Results were averaged on 10 subjects. An average classification accuracy of Ka is found to be 85% whereas the classification accuracy of Kha and Gha is in between 88% and 90%. The classification accuracy rate of Ga was found to be 78% which was lesser as compared to Kha and Gha.

Admissible wavelet packet sub‐band‐based harmonic energy features for Hindi phoneme recognition

In recent years, wavelet packet (WP) transform has been used as an important speech representation tool. WP-based acoustic features have found to be more effective than the short-time Fourier transform (STFT)-based features to capture the information of unvoiced phoneme in continuous speech. However, wavelet features fail to carry the same usefulness to represent the voiced phonemes such as vowels, nasals. This paper proposes new WP sub-band-based features by taking care of harmonic information of voiced speech signal. It has been noted that most of the voiced energy of the speech signal lies in between 250 and 2000 Hz. Thus, the proposed technique emphasises the individual sub-band harmonic energy up to 2 kHz. The speech signal is decomposed into 16 wavelet sub-bands and harmonic energy features are combined with WP cepstral (WPCC) features to enhance the performance of voiced phoneme recogniser. A standard phonetically balanced Hindi database is taken to analyse the performance of the proposed feature set. The noisy phoneme recognition task is also carried out to study the robustness. Significant improvement is obtained with the proposed feature set in voiced phoneme recognition over WPCC and conventional Mel frequency cepstral coefficient.

Development of Hindi speech stimuli to elicit auditory brainstem responses: Necessity and acoustic-phonetic considerations

Objective: Speech evoked auditory brainstem responses (spABRs) provide a window to understand the neurophysiological processing of speech in the brain. The temporal and spectral characteristics of speech sounds are faithfully represented in the subcortical responses; thus, these responses can be used to assess specific impairments in auditory processing. However, the spABRs are known to be affected by the acoustic properties of speech, language experience and training. Hence, the establishment of language specific speech stimuli is warranted for brainstem processing. The objective of the study was to develop Hindi speech stimuli for brainstem response recordings. Study design: The Hindi stop-voiced and voiceless phoneme of CV combination of 40 ms was synthesized. The behavioural identification score and discrimination ability of the selected stimuli were obtained from children (6–13 years) and adults (18–25 years) with normal hearing (NH) and moderate sensorineural hearing loss (HI). Brainstem evoked responses to stimulus |da| were gained from normal hearing and hearing impaired adults. Results: The identification and discrimination scores of stimuli between NH and HI children and adults were calculated. Statistically significant differences in the mean identification scores of synthesized speech stimuli were obtained. The mean, median, standard deviation, minimum, maximum and 95% confidence interval values of electrophysiological responses to speech stimulus were measured for the discrete peaks and V-A complex for both the groups. Conclusion: This article has delineated a comprehensive methodological approach for the development of Hindi speech stimuli and the recording of auditory brain stem responses to speech stimuli. The acoustic characteristic of the stimulus |da| was faithfully represented at brainstem level in normal hearing adults and found to be statistically different from HI individuals. This suggests that spABR offers an opportunity to segregate normal speech encoding from abnormal speech processing and implies that alterations in brainstem responses could have clinical significance for the identification of subjects with possible speech processing disorders.

Hindi phoneme classification using Wiener filtered wavelet packet decomposed periodic and aperiodic acoustic feature

Wavelet packet (WP) acoustic features are found to be very promising in unvoiced phoneme classification task but they are less effective to capture periodic information from voiced speech. This motivated us to develop a wavelet packet based feature extraction technique that signifies both the periodic and aperiodic information. This method is based on parallel distributed processing technique inspired by the human speech perception process. This front end feature processing technique employs Equivalent Rectangular Bandwidth (ERB) filter like wavelet speech feature extraction method called Wavelet ERB Sub-band based Periodicity and Aperiodicity Decomposition (WERB-SPADE). Winer filter is used at front end to minimize the noise for further processing. The speech signal is filtered by 24 band ERB like wavelet filter banks, and then the output of each sub-band is processed through comb filter. Each comb filter is designed individually for each sub-band to decompose the signal into periodic and aperiodic features. Thus it carries the periodic information without losing certain important information like formant transition incorporated in aperiodic features. Hindi phoneme classification experiments with a standard HMM recognizer under both clean-training and multi-training condition is conducted. This technique shows significant improvement in voiced phoneme class without affecting the performance of unvoiced phoneme class.

Neurophysiologic measures of auditory brainstem responses to Hindi speech stimulus in typically developing children

Context: Speech evoked auditory brain stem responses (spABR) are used to assess brain stem ability to encode speech. However, brain stem representations of speech sound are affected by acoustic differences of speech, language background and experiences. Hence, linguistically and culturally specific stimulus should be used. Therefore, there is need to investigate the brain stem encoding of Hindi speech sound. Objective: The purpose of the study was to investigate the spABR to Hindi stimulus in typically developing children. Methods and Material: Total 25 atypical children of the mean age of 6.3 years with SD 2.3 years in the age range of 8-12 years were recruited. A 40ms duration Hindi stop phoneme of CV combination |da| was synthesized to gain response of consonant (transient) and vowel (sustained) portion of speech stimulus from all subjects. Results: The spABR latencies and amplitudes of discrete peaks as well as the latency, amplitude, area & slope of V-A complex were measured. The stimulus to response timing and spectral magnitudes of periodic portion were also calculated. The mean, median, standard deviation, minimum, maximum and 95% confidence interval values were calculated. Conclusions: The obtained spABR values suggest faithful representation of acoustic characteristic of speech at brain stem level. These values are found to be similar with previous reports. Hence, spABR can serve as important tools in the diagnostic work up of auditory processing deficits in children to understand speech perception and production abilities.

Admissible wavelet packet features based on human inner ear frequency response for Hindi consonant recognition

It was observed that for non-stationary and quasi-stationary signals, wavelet transform has been found to be an effective tool for the time–frequency analysis. In the recent years wavelet transform being used for feature extraction in speech recognition applications. Here a new filter structure using admissible wavelet packet analysis is proposed for Hindi phoneme recognition. These filters have the benefit of having frequency bands spacing similar to the auditory Equivalent Rectangular Bandwidth (ERB) scale whose central frequencies are equally distributed along the frequency response of human cochlea. The phoneme recognition performance of proposed feature is compared with the standard baseline features and 24-band admissible wavelet packet-based features using a Hidden Markov Model (HMM) based classifier. Proposed feature shows better performance compared to conventional features for Hindi consonant recognition. To evaluate the robustness of proposed feature in the noisy environment NOISEX-92 database has been used.

Hindi phoneme-viseme recognition from continuous speech

Automatic Speech Recognition (ASR) system performs well under restricted conditions but the performance degrades under noisy environment. Audio–visual features play an important role in ASR systems in the presence of noise. In this paper, Hindi phoneme recognition system is designed using audio-visual features. The Discrete Cosine Transform (DCT) features of the lip region integrated with Mel Frequency Cepstral Coefficient (MFCC) audio features are used to get better recognition performance under noisy environments. Colour intensity, hybrid method and Pseudo-Hue methods have been used for lip-localisation approach with Linear Discriminant Analyser (LDA) as a classifier. Recognition performance using Pseudo-Hue method proved best among all the methods.

Automatic phonetic segmentation of Hindi speech using hidden Markov model

In this paper, we study the performance of baseline hidden Markov model (HMM) for segmentation of speech signals. It is applied on single-speaker segmentation task, using Hindi speech database. The automatic phoneme segmentation framework evolved imitates the human phoneme segmentation process. A set of 44 Hindi phonemes were chosen for the segmentation experiment, wherein we used continuous density hidden Markov model (CDHMM) with a mixture of Gaussian distribution. The left-to-right topology with no skip states has been selected as it is effective in speech recognition due to its consistency with the natural way of articulating the spoken words. This system accepts speech utterances along with their orthographic “transcriptions” and generates segmentation information of the speech. This corpus was used to develop context-independent hidden Markov models (HMMs) for each of the Hindi phonemes. The system was trained using numerous sentences that are relevant to provide information to the passengers of the Metro Rail. The system was validated against a few manually segmented speech utterances. The evaluation of the experiments shows that the best performance is obtained by using a combination of two Gaussians mixtures and five HMM states. A category-wise phoneme error analysis has been performed, and the performance of the phonetic segmentation has been reported. The modeling of HMMs has been implemented using Microsoft Visual Studio 2005 (C++), and the system is designed to work on Windows operating system. The goal of this study is automatic segmentation of speech at phonetic level.

WAVELET SUB-BAND BASED TEMPORAL FEATURES FOR ROBUST HINDI PHONEME RECOGNITION

This paper proposes the use of wavelet transform-based feature extraction technique for Hindi speech recognition application. The new proposed features take into account temporal as well as frequency band energy variations for the task of Hindi phoneme recognition. The recognition performance achieved by the proposed features is compared with the standard MFCC and 24-band admissible wavelet packet-based features using a linear discriminant function based classifier. To evaluate robustness of these features, the NOISEX database is used to add different types of noise into phonemes to achieve signal-to-noise ratios in the range of 20 dB to -5 dB. The recognition results show that under noisy background the proposed technique always achieves a better performance over MFCC-based features.

Implicit learning of non‐native speech stimuli.

Previous studies have induced robust adult learning of non‐native natural speech stimuli using explicit instructions and feedback. Here, Greek adults were exposed to non‐native Hindi phonemes for six sessions in which they were unaware of the phoneme distinctions and the phonemes had no relevance to their main task. Stimuli were natural recordings of CV syllables (a retroflex or dental unvoiced stop followed by [a]) from a Hindi speaker. On each trial, participants heard two identical dental sounds and two retroflex sounds differing in intensity by an adaptively varied amount between 0.5–3 dB. The explicit task was to identify which pair differed in intensity, while the implicit learning goal was the phonetic contrast. After training, participants were tested on explicit phonetic identification and discrimination of trained and untrained stimuli spoken by the same and a novel speaker. The trained group outperformed an untrained group who were administered the same tests. Our results are compatible with a perceptual learning model where task‐irrelevant stimuli (Hindi sounds) are learned due to a systematic pairing with task targets (intensity difference). Learning did not extend to the novel speaker, in agreement with previous studies showing that variability in training is crucial for generalization.