S. S. Agrawal

Dialectal influences on acoustic duration of Hindi phonemes

In this paper, the authors have explored the influence of dialects on the prosodic feature represented by duration of phonemes in spoken utterances. A dialect is any distinguishable variety of a language spoken by a group of people. Dialect of a given language are the differences in speaking styles of a particular language, because of geographical and ethnic differences of the speakers. The pronunciation and choice of words during communication by individuals are governed by their exposure to native languages. Hindi is the first language for communication in India and is spoken in ten of its states. The spoken form of Hindi in different regions of these states vary from one another. These variations highly influence the prosodic structure of phonemes. In this work, six different dialects of Hindi namely, Awadhi, Bagheli, Bhojpuri, Bundeli, Haryanvi and Khariboli are considered for studying their influence on phoneme duration in spoken utterances. All attempts have been made in this study to analyze and describe the concrete correlation between phoneme lengthening in different dialects based on its position in the word at start and end. Observations show that duration of semi vowels and dental phonemes can be used for identification of Bagheli dialect, Awadhi and Bhojpuri dialects can be identified using duration of palatal sounds. Duration of velar sounds and fricatives gives clues for identification of Khariboli dialect. From the results obtained it has been analyzed that phoneme duration is an important parameter for dialect identification.

Fusion of multi-stream speech features for dialect classification

Current research in the area of voice recognition has entered a new stage. It does not only concentrate on the correct evaluation of linguistic information embodied in the speech signal, it also works towards identification of variations naturally present in speech. Undoubtedly, the focus is to enhance the accuracy and precision of the developed technique. Speaker’s accent due to his native dialect is one of the major source of variability. Prior knowledge of the spoken dialect will help in the creation of multi-model speech recognition system and can enhance its recognition performance. This paper focusses on applying some of the established dialect identification techniques to identify speaker’s spoken dialect among dialects of Hindi. Fusion of multiple streams obtained as a combination of phonetic and prosodic features is implemented to exploit the acoustic information. The work presented here also exploits the ability of AANN to capture the distribution of data points in a reduced number and further to classify them into groups. System performance for different level of fusion is recorded for Hindi dialect classification. It is observed that Duration as prosodic feature is an important cue for automatic dialect identification systems.

Continuous density Hidden Markov Model for context dependent Hindi speech recognition

With the advancement in technology and the inherent advantage of voice based communication due to its variability, speed and convenience has driven attention towards mechanical recognition of speech. Literature survey of research in this area shows that almost every system uses Gaussian Mixture Hidden Markov model for optimal performance in recognition of speech. Evaluation of Gaussian likelihood dominates the total computational load in using this statistical approach. The appropriate selection of Gaussian mixture is very important. Current choice of mixture component is arbitrary with little justification. Also the standard set for European languages can not be used in Hindi speech recognition due to mismatch in database size of the languages. Parameter estimation with too many or few component may inappropriately estimate the mixture model. Therefore, number of mixture is important for expectation maximization process. In this research work, the authors estimate number of Gaussian mixture component for Hindi database based upon the size of vocabulary. MFCC and PLP features along with its extended version has been used as speech feature. HLDA is applied for feature reduction while using extended features.

Development of Hindi mobile communication text and speech corpus

This paper describes the collection of a text and audio corpus for mobile personal communication in Hindi. Hindi is the largest of the Indian languages, and is the first language for more than 200 million people who use it not only for spoken mobile communication but also for sending text messages to each other. The main script for Hindi is Devanagari, but it is not well supported by the current generation of mobile devices. The Devanagari alphabet is twice as large as for English which makes it difficult to fit onto the small keypad of a mobile device. The aim of this project is to collect text and speech resources which can be used for training spoken language systems that aide text messaging on mobile devices - i.e. train a speech recogniser for the mobile personal communication domain so that text can be input through dictation rather than by typing. In total we collected a text corpus of 2 million words of natural messages in 12 different domains, and a spoken corpus of 100 speakers who each spoke 630 phonetically rich sentences - about 4 hours of speech. The speech utterances were recorded in 16 kHz through 3 recording channels: a mobile phone, a headset and a desktop mounted microphone. The data sets were properly annotated and available for development of speech recognition / synthesis systems in mobile domain.

Advances in Voice Enabled Human Machine Communication

inherent advantage of speech communication due to its variability, convenience and speed along with our increasing requirements to communicate with machines has driven the attention of researchers towards mechanical recognition of speech. Technological advancements and improvements in the fundamental approaches have shown a successful transition from small vocabulary isolated word recognition to large vocabulary continuous speech recognition. Even after years of research and development the accuracy of automatic speech recognition remains one of the major challenges. Design of speech recognition system requires careful selection of feature extraction technique and modeling approach to cover the challenges faced due to variability of speech-speaker characteristic, storage space and processing speed requirements. In this paper an effort has been made to highlight the progress made so far for mechanizing the recognition of speech along with the major challenges in this field. Authors have also presented a brief description of voice enabled service for common people. The objective of this paper is to summarize some of the well known methods used at various stages of speech recognition system along with their benefits and limitations.

Speaker-Independent Recognition System for Continuous Hindi Speech Using Probabilistic Model

In this generation of IT, communicating with machines in an expedient manner using human speech that too in their own language is highly desirable. This is achieved using speech recognition systems which allow the general public to speak to the machine by recognizing their voice. Hindi being the most widely spoken language with approx. 260 million first-language speakers [1] should have a real-time recognition system. The main objective of this paper is to develop a speaker-independent system which can recognize continuous Hindi speech in real-time scenario. This paper presents the feasibility of MFCC for feature extraction and dynamic time warping to compare the test sequence. The system has been trained on 8 h of audio data and a trigram language model trained with 30K words. With a dictionary of 6K words, the system gives a word accuracy of 80–85%.