Hindi Speech Research Articles

A Hindi Speech Actuated Computer Interface for Web Search

Aiming at increasing system simplicity and flexibility, an audio evoked based system was developed by integrating simplified headphone and user-friendly software design. This paper describes a Hindi Speech Actuated Computer Interface for Web search (HSACIWS), which accepts spoken queries in Hindi language and provides the search result on the screen. This system recognizes spoken queries by large vocabulary continuous speech recognition (LVCSR), retrieves relevant document by text retrieval, and provides the search result on the Web by the integration of the Web and the voice systems. The LVCSR in this system showed enough performance levels for speech with acoustic and language models derived from a query corpus with target contents.

Preface to the special issue on “Signal Processing Applications in Human Computer Interaction”

1. Transformation of Whispering Voice to Pseudo-Real Voice for Unvoiced Telephony and Communication aid for Voice-Handicapped Persons Anderson Pierre Passos 2. Agent-based Framework for Intelligent Natural Language Interface Moses Ekpenyong and Eno-Abasi Urua 3. An Efficient Ear Recognition Technique Invariant to Illumination and Pose Surya Prakash and Phalguni Gupta 4. Ultrasonic Digital Signal Processing Simulation in Viscoelastic Medium with Generalized Parametric Function R.M. Farouk and M.G.S. Ali 5. Performance Evaluation of Sequentially Combined Heterogeneous Feature Streams for Hindi Speech Recognition System Rajesh Kumar Aggarwal and M. Dave 6. Speech Emotion Recognition Approaches in Human Computer Interaction S. Ramakrishnan and Ibrahiem M.M. El Emary 7. Implementation of a Robust Absolute Virtual Head Mouse Combining Face Detection, Template Matching and Optical Flow Algorithms

Performance evaluation of sequentially combined heterogeneous feature streams for Hindi speech recognition system

State-of-the-art automatic speech recognition (ASR) systems follow a well established statistical paradigm, that of parameterization of speech signals (a.k.a. feature extraction) at front-end and likelihood evaluation of feature vectors at back-end. For feature extraction, Mel-frequency cepstral coefficients (MFCC) and perceptual linear prediction (PLP) are the two dominant signal processing methods, which have been used mainly in ASR. Although the effects of both techniques have been analyzed individually, it is not known whether any combination of the two can produce an improvement in the recognition accuracy or not. This paper presents an investigation on the possibility to integrate different types of features such as MFCC, PLP and gravity centroids to improve the performance of ASR in the context of Hindi language. Our experimental results show a significant improvement in case of such few combinations when applied to medium size lexicons in typical field conditions.

Transformation of Emotion Based on Acoustic Features of Intonation Patterns for Hindi Speech and their Perception

Changes in intonation patterns may convey not only different meaning, but also different emotions even if the sequence of speech segments is the same in a sentence. The patterns change depending upon the structure and emotion of the sentence and require being stored in speech database. It is a difficult and time-consuming task to store all utterances of all the expressive style, which also consumes huge memory space. Therefore, there should be an approach that minimizes the time and memory space for emotion-rich database. A number of studies in this respect have been carried out for several languages and models developed. However, for Hindi not many studies have been carried out. Taking this fact into consideration, the intonation patterns have been studied and analysed for Hindi language and compared with some other languages. On the basis of the analysis on intonation patterns, an algorithm has been proposed for emotion conversion. This algorithm requires storing of neutral utterances only in the database and other expressive style utterances can be derived from these neutral utterances. The proposed algorithm is based on the linear modification model, where fundamental frequency is a major factor to convert emotions. In this paper, surprise, sadness, happiness, and anger are taken as target emotions. For performing experiments and verification of results, an emotion-rich speech database is prepared with the help of 15 native speakers. Each speaker has been given 20 sentences and asked to generate utterances in five expressive states “neutral”, “sadness”, “anger”, “surprise”, and “happiness”. These sentences are recorded in normal laboratory conditions with a 44.1 KHz sampling rate and 16-bit precision with mono channel. In the perception test, group of listeners were asked to listen to the utterances from database and judge the emotion. This perception test involves classification of the emotions already available in the database by the listener and to judge the quality of converted neutral utterances. The results are analysed and performance of the experiment is evaluated. The accuracy of the perception test on transformed emotions was found out to be 93.2% for surprise, 91.6% for sadness, 83% for happiness, and 95.3% for anger.

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.

Exploring the Discrete Wavelet Transform as a Tool for Hindi Speech Recognition

Exploring the Discrete Wavelet Transform as a Tool for Hindi Speech Recognition

Optimum HMM combined with vector Quantization for Hindi Speech word Recognition

This paper proposes an optimum speaker-independent, isolated word Hidden Markov Model (HMM) recognizer for the Hindi language. The recognition system is based on the combination of the vector quantization (VQ) technique at the acoustical level and the Markovian modeling at the recognition level. The recognizer consists of three modules – feature extraction, vector quantizer and HMM training and testing modules. The scheme proposed here firstly computes the acoustic features in terms of the Linear Predictive Cepstral LPC coefficients, Mel-Frequency Cepstral coefficients and delta MFCC along with noise and silence detection. Then, codebooks are created using VQ, and finally in the recognition phase, an optimum set of parameters are derived from different phases for getting the highest recognition score. The training and testing database consists of a set of 35 utterances of nine Indian cities/states and 35 utterances of nine digits spoken in Hindi by male and female speakers. The recognition rate was observed to be 98.61%.

Hybrid wavelet based LPC features for Hindi speech recognition

Hybrid features are presented for speech recognition that uses linear prediction in combination with multi-resolution capabilities of wavelet transform. Wavelet-Based Linear Prediction Coefficients (WBLPC) are obtained by applying 3 and 4-level wavelet decomposition and then having linear prediction of each sub-bands to get total 13 features. These features have been tested using a linear discriminant function and Hidden Markov Model (HMM) based classifier for speaker dependent and independent isolated Hindi digits recognition. 3-level WBLPC features gave higher percentage recognition than LPC features while 4-level WBLPC features using HMM gave the highest percentage recognition for both speaker dependent and independent cases.

The effect of a hearing aid noise reduction algorithm on the acquisition of novel speech contrasts

Audiologists are reluctant to prescribe digital hearing aids with active digital noise reduction (DNR) to pre-verbal children due to their potential for an adverse effect on the acquisition of language. The present study investigated the relation between DNR and language acquisition by modeling pre-verbal language acquisition using adult listeners presented with a non-native speech contrast. Two groups of normal-hearing, monolingual Anglophone subjects were trained over four testing sessions to discriminate novel, difficult to discriminate, non-native Hindi speech contrasts in continuous noise, where one group listened to both speech items and noise processed with DNR, and where the other group listened to unprocessed speech in noise. Results did not reveal a significant difference in performance between groups across testing sessions. A significant learning effect was noted for both groups between the first and second testing sessions only. Overall, DNR does not appear to enhance or impair the acquisition of novel speech contrasts by adult listeners.

A large-vocabulary continuous speech recognition system for Hindi

In this paper we present two new techniques that have been used to build a large-vocabulary continuous Hindi speech recognition system. We present a technique for fast bootstrapping of initial phone models of a new language. The training data for the new language is aligned using an existing speech recognition engine for another language. This aligned data is used to obtain the initial acoustic models for the phones of the new language. Following this approach requires less training data. We also present a technique for generating baseforms (phonetic spellings) for phonetic languages such as Hindi. As is inherent in phonetic languages, rules generally capture the mapping of spelling to phonemes very well. However, deep linguistic knowledge is required to write all possible rules, and there are some ambiguities in the language that are difficult to capture with rules. On the other hand, pure statistical techniques for baseform generation require large amounts of training data, which is not readily available. We propose a hybrid approach that combines rule-based and statistical approaches in a two-step fashion. We evaluate the performance of the proposed approaches through various phonetic classification and recognition experiments.

A large-vocabulary continuous speech recognition system for Hindi

In this paper we present two new techniques that have been used to build a large-vocabulary continuous Hindi speech recognition system. We present a technique for fast bootstrapping of initial phon...

Cross-language speech perception: Evidence for perceptual reorganization during the first year of life

Previous work in which we compared English infants, English adults, and Hindi adults on their ability to discriminate two pairs of Hindi (non-English) speech contrasts has indicated that infants discriminate speech sounds according to phonetic category without prior specific language experience (Werker, Gilbert, Humphrey, & Tees, 1981), whereas adults and children as young as age 4 (Werker & Tees, in press), may lose this ability as a function of age and or linguistic experience. The present work was designed to (a) determine the generalizability of such a decline by comparing adult English, adult Salish, and English infant subjects on their perception of a new non-English (Salish) speech contrast, and (b) delineate the time course of the developmental decline in this ability. The results of these experiments replicate our original findings by showing that infants can discriminate non-native speech contrasts without relevant experience, and that there is a decline in this ability during ontogeny. Furthermore, data from both cross-sectional and longitudinal studies shows that this decline occurs within the first year of life, and that it is a function of specific language experience.

Some patterns of unscripted speech in Hindi

This paper presents data from a small corpus of unscripted speech gathered from one male and one female adult native speaker of Hindi. An acoustic analysis of the data demonstrated changes such as lenition and assimilation. The cases of assimilation included stop plus stop sequences yielding geminates. The results of a perceptual test showed that such ‘pseudogeminates’ are generally perceived as true geminates. Parallels between these phenomena and historical sound changes in Indo-Aryan are also discussed.

Investigating a computer-based method to measure school-age children’s ability to discriminate non-native speech contrasts

Most studies of non-native speech perception have examined adult subject populations, as tests on infants, toddlers, and older children can be methodologically difficult to conduct. Direct application of methods used for adults in the assessment of children can be problematic. Such methods can be long, mentally fatiguing, and cognitively complex. In addition to task variables, stimuli which are difficult to perceive, such as non-native speech sounds, make such tests on children all the more difficult. The present study examines a new computer-based testing method specifically designed to test the discrimination of non-native speech sounds with 5- to 15-year-old children. The design uses an oddity task that presents non-native phoneme contrasts in a way that is engaging and motivating. For baseline comparison, discrimination data were obtained from 10 AE and 10 Hindi speaking adults using four phonetically different, natural CV Hindi speech contrasts and compared to identification data previously obtained from AE and Hindi adults using the same stimuli (Pruitt, 1995) implemented with a standard adult-testing protocol. Results from pilot testing with children and adults using this new computer-based method will be discussed and compared to the baseline identification and discrimination data. [Work supported by NICHD HD37954 to P. K. Kuhl.]

Word boundary hypothesization for continuous speech in Hindi based on F 0 patterns

This paper proposes an algorithm based on F 0 patterns to hypothesize word boundaries and function words in continuous speech in Hindi. It makes use of the properties of F 0 contour such as declination tendency, resetting and fall-rise patterns in Hindi. The syllabic units are identified by using the energy contour, pitch and the first order LP coefficient. Each syllabic unit is assigned an accent value L (Low), H or h (High) by (i) comparing the F 0 value at the mid point of each syllabic nucleus with that of the previous syllabic unit and (ii) comparing the F 0 values at two different points within each syllabic unit in a sequence having an accent value L. Word boundaries are placed between the adjacent syllabic units (i) H and L, (ii) h and L, (iii) L and L, (iv) L and h and (v) H and h. An evaluation conducted on a corpus of 50 sentences in Hindi read aloud by five native speakers in an ordinary office environment showed that about 74 percent of the word boundaries and about 28 percent of the function words were correctly identified. The results of the word boundary hypothesization can be used to improve the performance of the acoustic-phonetic, lexical and syntactic modules in a speech-to-text conversion system. Robustness of the algorithm in handling noisy speech input conditions and telephone speech are also discussed.

Intonation component of a text-to-speech system for Hindi

In this paper, we describe some features of the fundamental frequency (F 0) contours of speech in Hindi and propose an approach to represent and activate this intonation knowledge for an unrestricted text-to-speech system for Hindi. As in most languages, the declarative sentences in Hindi show a declining pattern of F 0 contour, whereas interrogative sentences show a rising F 0 contour. The backdrop declining or rising pattern is characterized by local falls and rises which are determined by the phonological pattern of the constituent words. In complex declarative sentences the F 0 contour resets at major syntactic boundaries. Experiments to obtain the inherent F 0 of segments are described. The intonation knowledge derived from analysis of speech is coded in a production system format. Intelligibility and naturalness of the synthesized speech improved significantly after incorporation of the intonation rules.

Use of fuzzy mathematical concepts in character spotting for automatic recognition of continuous speech in Hindi

In this paper we describe the use of fuzzy mathematical concepts in spotting characters from continuous speech in the Indian language Hindi. The research effort reported here is concerned with identifying the phonetic features from the acoustic parameters of a speech signal and combining the features to spot a character in continuous speech. These functions are performed by using expert systems that use confidence values for grading the conclusions arrived at each stage of processing. This paper discusses how this grading is obtained by using numerical translation of the description of a character obtained from an expert phonetician. Confidence values are assigned to the premises of the rules obtained from an expert phonetician using fuzzy membership functions. Fuzzy relations are used to grade the conclusions derived from these premises. Results of applying these techniques to character spotting are given.

Word boundary hypothesization in Hindi speech

This paper proposes a method for hypothesizing word boundaries in Hindi speech. The method is based on the observation that function words such as case markers, pronouns and conjunctions occur frequently in Hindi text and spotting of these frequently occurring patterns is proposed as a means for hypothesizing word boundaries in a speech-to-text conversion system for Hindi. Initially, the idea was tested on a correct text with all word boundaries (except sentence boundaries) removed; the results showed that nearly 67% of the word boundaries were correctly hypothesized. Later, experiments with input containing errors simulated to represent speech environment showed that the proposed method is effective even at error levels as high as 50%. The implications of these results in the development of a speech-to-text conversion system for Hindi are discussed.

Cross-language speech perception: Evidence for perceptual reorganization during the first year of life

Previous work in which we compared English infants, English adults, and Hindi adults on their ability to discriminate two pairs of Hindi (non-English) speech contrasts has indicated that infants discriminate speech sounds according to phonetic category without prior specific language experience (Werker, Gilbert, Humphrey, & Tees, 1981), whereas adults and children as young as age 4 (Werker & Tees, in press), may lose this ability as a function of age and or linguistic experience. The present work was designed to (a) determine the generalizability of such a decline by comparing adult English, adult Salish, and English infant subjects on their perception of a new non-English (Salish) speech contrast, and (b) delineate the time course of the developmental decline in this ability. The results of these experiments replicate our original findings by showing that infants can discriminate non-native speech contrasts without relevant experience, and that there is a decline in this ability during ontogeny. Furthermore, data from both cross-sectional and longitudinal studies shows that this decline occurs within the first year of life, and that it is a function of specific language experience. © 2002 Published by Elsevier Science Inc.

Developmental Aspects of Cross-Language Speech Perception

Previous research has suggested that infants discriminate many speech sounds according to phonemic category regardless of language exposure, while adults of one language group may have difficulty discriminating nonnative linguistic contrasts. Our study attempted to address directly questions about infant perceptual ability and the possibility of its decline as a function of development in the absence of specific experience by comparing English-speaking adults, Hindi-speaking adults, and 7-month-old infants on their ability to discriminate 2 pairs of natural Hindi (non-English) speech contrasts. To do this, infants were tested in a "visually reinforced infant speech discrimination" paradigm, while a variant of this paradigm was used to test adults. Support was obtained for the above hypotheses. Infants were shown to be able to discriminate both Hindi sound pairs, and support for the idea of a decrease in speech perceptual abilities wih age and experience was clearly evident with the rarer of the 2 non-English contrasts. The results were then discussed with respect to the possible nature and purpose of these abilities.