Rajkishore Prasad

Robots That Can Hear, Understand and Talk

In this survey paper we analytically examine the state of the art in speech and natural language processing technologies, and one of their most promising applications in the robotics world as a user interface to facilitate human-robot interaction/communication and robot control by spoken natural language. Theoretical aspects of spoken language technology and the main bottlenecks in developing a conversational interface for a robot have been presented in depth with results found while searching the literature related to the major breakthroughs made in this field. In this study, we present a brief technical introduction to talk-active robots, and to discuss related future technical challenges and technical approaches used. Efforts have been made to highlight the limitations and missing directions of the research and development in the spoken language technology which are creating hurdles in the development of voice-active robots for real-world applications.

Estimation of Shape Parameter of GGD Function by Negentropy Matching

In this paper we present a novel method for the estimation of the shape parameter of the Generalized Gaussian Distribution (GGD) function for the leptokurtic and Gaussian signals by matching negentropy of GGD function and that of data approximated by some non-polynomial functions. The negentropy of GGD function is monotonic function of its shape parameter for values corresponding to super-Gaussian and Gaussian distribution family. The simulation results have been compared with those obtained by existing methods such as Mallat’s method and Kurtosis matching method. It has been found that the proposed method is effective and useful in the cases where we have a few observation samples and distribution is highly spiky.

Blind Separation of Speech by Fixed-Point ICA with Source Adaptive Negentropy Approximation

This paper presents a study on the blind separation of a convoluted mixture of speech signals using Frequency Domain Independent Component Analysis (FDICA) algorithm based on the negentropy maximization of Time Frequency Series of Speech (TFSS). The comparative studies on the negentropy approximation of TFSS using generalized Higher Order Statistics (HOS) of different nonquadratic, nonlinear functions are presented. A new nonlinear function based on the statistical modeling of TFSS by exponential power functions has also been proposed. The estimation of standard error and bias, obtained using the sequential delete-one jackknifing method, in the approximation of negentropy of TFSS by different nonlinear functions along with their signal separation performance indicate the superlative power of the exponential-power-based nonlinear function. The proposed nonlinear function has been found to speed-up convergence with slight improvement in the separation quality under reverberant conditions.

Enhancement of speech signals separated from their convolutive mixture by FDICA algorithm

This paper presents a novel method for the enhancement of independent components of mixed speech signal segregated by the frequency domain independent component analysis (FDICA) algorithm. The enhancement algorithm proposed here is based on maximum a posteriori (MAP) estimation of the speech spectral components using generalized Gaussian distribution (GGD) function as the statistical model for the time-frequency series of speech (TFSS) signal. The proposed MAP estimator has been used and evaluated as the post-processing stage for the separation of convolutive mixture of speech signals by the fixed-point FDICA algorithm. It has been found that the combination of separation algorithm with the proposed enhancement algorithm provides better separation performance under both the reverberant and non-reverberant conditions.

Single Channel Speech Enhancement: MAP Estimation Using GGD Prior Under Blind Setup

This paper presents a statistical algorithm using Maximum A Posteriori (MAP) estimation for the enhancement of single channel speech, contaminated by the additive noise, under the blind framework. The algorithm uses Generalized Gaussian Distribution (GGD) function as a prior probability to model magnitude of the Spectral Components (SC) of the speech and noise in the frequency domain. An estimation rule has been derived for the estimation of the SC of the clean speech signal under the presence of additive noise signal. Since the parsimony of the GGD distribution depends on its shape parameter, it provides flexible statistical model for the data with different distribution, e.g. impulsive, Laplacian, Gaussian, etc. The enhancement result for Laplacian noise have been presented and compared with that of the conventional Wiener filtering, which assumes Gaussian distribution for SCs of both the speech and noise.

hummgenic changes in large scale temporal correlation of EEG in BP

In this paper we present results of detrended fluctuation analysis (DFA) on raw EEG data obtained from subjects performing Bhramari Pranayama (BP).BP is characterized by the production of low frequency humming sound like that of bumble bee by sustaining pronunciation of nasal /m/ sound while keeping ears occluded and oral cavity closed at lips. BP is effective in healing many neuronal disorders. We hypothesize that the humming sound is playing such healing role by changing brain wave patterns. The results of DFA analysis show that BP changes scaling exponents for the raw EEG data in the frontal and temporal region of the brain. Decrement in scaling exponent lowers decay of temporal correlations in data while increment in scaling exponent provides rapid decay. The estimated exponent were found to lie between 0.5 and 1.6 which also show that EEG signals are generated through fractal process and contain long range temporal correlates. It is hypothesized that the interaction of self humming and brain wave signals might be actor behind bringing good effects by BP.

Changes in auditory threshold of hearing after Bhramari Pranayama

This paper presents changes in relative auditory sensitivity in human after a yogic practice known as Bhramari Pranayama (BP). BP is a breathing exercise and is helpful in reducing neuronal abnormalities such as stress, hypertension etc. BP is characterized by the humming sound, one has to produce while doing it. The produced humming sound is steady and low frequency sound like the humming sound of the bumble bee. We assume that the self humming sound brings many soothing effects for the subject doing it. This paper reports effect of such humming sound on the Auditory Threshold of Hearing (ATH), a measure of auditory sensitivity. The observations on plural number of subjects show that BP produces change in ATH level for frequency range 100 Hz-15 kHz, however, such change last for a few minutes and ATH level again restores to normal level. We hope, the observed change in sensitivity might be beneficial for the ear functioning as such changes are related to cochlear mechanics.

how to hum like a bumble BEE

This paper is a part of scientific investigations on Bhramari Pranayama (BP), a yogic breathing exercise in which practitioners imitate humming sound of the bumble bee while exhaling strictly through nostrils. It is still undiscovered why the inventor (ancient Indian Yoga masters) instructed to hum like a bumble bee and not like a common bee or other insects. It is hypothesized that it is the humming sound and humgenic vibrations that are playing key role in bringing many of the claimed benefits of BP. The sound production system of the bumble bee and that of humans are different, however, due to inbuilt complexities and articulatory capacity humans sound production system can produce variety of sound. Humans can produce humming sound like a bumble bee by sustaining pronunciations of nasal consonants /m/ or /n/ or /ng/ alone or as an aid for continuation of nasalization of vowels or others. The results presented in this paper show sustained pronunciation of /m/ is easier and very similar to that of bumble bee sound in the spectral content. The old texts on Yoga also suggest humming by sustaining /m/ part of the /om/ sound.