Research Areas:

  • Speech Processing
  • Speech Recognition
  • Speech Enhancement
  • Digital Signal Processing
  • Human-Computer Interface

 

 

Novel methods developed by me:

Other Research Interests:

  • Edge Detection and Image Segmentation
  • Expert Systems
  • Neural Networks and Fuzzy Systems

 

    Research Projects 

April 01 ~ August 31       2007

JSPS Post-Doctoral Research Fellowship, Japan.

(The Ministry of Education, Culture and Sports, Japan)

 

 

Title: Study on canonicalization for robust speech recognition.

Summary:

Canonicalization is a process for eliminating the influence of various hidden factors in automatic speech recognition (ASR) that decrease acoustic likelihood between a targeting category and the other categories. The hidden factors include acoustic transfer function, gender, noise types and their SN ratios, age (child/old), speaking style, phone context, and so on. The proposed canonicalization process maps an input acoustic feature onto a DPF space, which is invariant to acoustic environments, widens the margin of the acoustic likelihood, which results in less number of mixtures in HMM, and consequently gives a functionality of a high performance phonetic typewriter .

To realize the canonicalization process, we simulate a human auditory system at the upper level of the front-end in our ASR system. The goal of the research was to solve the problem of coarticulation of phonemes at their transitions and thereby increase recognition performance.

 

December 2008

Submitted

CCIS, King Saud University.

 

Title: Pitch Detection in Noisy Environments.

Summary:

Precise calculation of pitch in the speech signal has demonstrated to be a basic task in almost all areas of speech research. Consequently, a wide range of perceptual models and algorithms using a variety of techniques and a varying degree of accuracy to extract pitch exist. However, the pitch detection algorithms (PDAs) face a real challenge in presence of noise. Although some noise-robust PDAs are reported in recent publications, they suffer from huge computational complexity or vulnerable to severe noisy conditions.

The objective of the current research is to develop a noise-robust, real time PDA, which is easy to integrate with other applications.

 

December 2008

Submitted

CCIS, King Saud University.

 

Title: Computationally Efficient Intelligent Module for Computer Game Engine.

Summary:

The fundamental problem to create realistic scenes in computer games is that realism requires complex and highly detailed models. Complexity makes things run slowly especially when we want rapid, and sometimes interactive, rendering and animation. This imposes restrictions on a game engine in allotting much of a CPU time for an intelligent module. As a result there is a need for a computationally efficient intelligent module which can handle input voice commands with minimal time effort leaving most of a CPU time for graphics rendering.

The goal is to recognize any arbitrary input commands in English words using short interaction without human effort.