PGA implementation of speech recognition system based on HMM

Dissertant

al-Saeedy, Ala Abd al-Husayn Refeis

Thesis advisor

Abbas, Iyad Ibrahim

University

University of Technology

Faculty

-

Department

Department of Electrical Engineering

University Country

Iraq

Degree

Master

Degree Date

2012

English Abstract

This thesis introduced an approach to design and implement an embedded SoPC (System on Programmable Chip) technique with Altera Nios II processor on a FPGA chip for real-time speech recognition system by developing hardware/software with minimum usage of resources (hardware components) and relatively small size software.

This reduces the memory utilization, achieved by using Mel Frequency Cepstral Coefficients (MFCCs) technique as feature extraction combined with its first derivative (AMFCCs) including power computation of the speech frames (i.ى, MFCC, AE, andAMFCC) , called observation vector of the speech signal.

To model the obtained observation, Gaussian Mixture Model (GMM) has been used, which is passed to a Hidden Markov Model (HMM) as probabilistic model to process the GMM statistically to take a decision on the uttered words recognition, whether a single or composite, one or more syllable words (i.e.

one, six, welcome).

The words that are used for training and testing the system included selected English and Arabic words.

The framework was implemented on Cyclone II EP2C70F896C6N FPGA chip placed on ALTERA DE2-70 Development Board.

The utilities programs which are used as tools for designing and developing hardware/software are Quartus II 11.0sp1 (32-Bit) and Nios II 11.0sp1 IDE/C++ respectively.

Each word model is stored as Transition Matrix, Diagonal Covariance Matrices, and Mean Vectors in the memory and utilized only 4.45Kbytes to each word model regardless of the uttered word length.

Recognition rate gives as 100% for the already trained speakers.

The test was conducted at different sound levels of the surrounding environment (53dB to 73dB) as measured by Sound Level Meter (SLM) instrument.

Training and recognition software had size equal to 329Kbytes includes code and initialized data.

Main Subjects

Electronic engineering

• APA
• MLA
• AMA

Language

English

Data Type

Arab Theses

Record ID

BIM-418649