Experiments were carried out for continuous real speech mixed with either competitive speech source or broadband noise. Results show that separation outputs bear similar spectral trajectories as the ideal source signals. For speech mixtures, the proposed algorithm is evaluated in two ways: segmental signal-to-interference ratio (segSIR) and Itakura-Saito distortion ( dIS). It is found that (1) interference signal power is reduced in term of segSIR improvement, even under harsh condition of similar target speech and interference powers; and (2) dIS between the estimated source and the clean speech source is significantly smaller than before processing. These assert the capability of the proposed algorithm to extract individual sources from a mixture signal by reducing the interference signal and generating appropriate spectral trajectory for individual source estimates. / Our approach is based on the findings of psychoacoustics. To separate individual sound sources in a mixture signal, human exploits perceptual cues like harmonicity, continuity, context information and prior knowledge of familiar auditory patterns. Furthermore, the application of prior knowledge of speech for top-down separation (called schema-based grouping) is found to be powerful, yet unexplored. In this thesis, a bi-directional, model-based speech separation and enhancement algorithm is proposed by utilizing speech schemas, in particular. As model patterns are employed to generate subsequent spectral envelopes in an utterance, output speech is expected to be natural and intelligible. / The proposed separation algorithm regenerates a target speech source by working out the corresponding spectral envelope and harmonic structure. In the first stage, an optimal sequence of Wiener filtering is determined for subsequent interference removal. Specifically, acoustic models of speech schemas represented by possible line spectrum pair (LSP) patterns, are manipulated to match the input mixture and the given transcription if available, in a top-down manner. Specific LSP patterns are retrieved to constitute a spectral evolution that synchronizes with the target speech source. With this evolution, the mixture spectrum is then filtered to approximate the target source in an appropriate signal level. In the second stage, irrelevant harmonic structure from interfering sources is eliminated by comb filtering. These filters are designed according to the results of pitch tracking. / This thesis focuses on speech source separation problem in a single-microphone scenario. Possible applications of speech separation include recognition, auditory prostheses and surveillance systems. Sound signals typically reach our ears as a mixture of desired signals, other competing sounds and background noise. Example scenarios are talking with someone in crowd with other people speaking or listening to an orchestra with a number of instruments playing concurrently. These sounds are often overlapped in time and frequency. While human attends to individual sources remarkably well under these adverse conditions even with a single ear, the performance of most speech processing system is easily degraded. Therefore, modeling how human auditory system performs is one viable way to extract target speech sources from the mixture before any vulnerable processes. / Lee, Siu Wa. / "April 2008." / Adviser: Chung Ching. / Source: Dissertation Abstracts International, Volume: 70-03, Section: B, page: 1846. / Thesis (Ph.D.)--Chinese University of Hong Kong, 2008. / Includes bibliographical references (p. 233-252). / Electronic reproduction. Hong Kong : Chinese University of Hong Kong, [2012] System requirements: Adobe Acrobat Reader. Available via World Wide Web. / Electronic reproduction. [Ann Arbor, MI] : ProQuest Information and Learning, [200-] System requirements: Adobe Acrobat Reader. Available via World Wide Web. / Abstracts in English and Chinese. / School code: 1307.

This paper describes an isolated-word and speaker-independent Mandarin digit speech recognition system based on Backpropagation Neural Networks(BPNN). The recognition rate will achieve up to 95%. When the system was applied to a new user with adaptive modification method, the recognition rate will be higher than 99%. In order to implement the speech recognition system on Digital Signal Processors (DSP) we use a neuron-cancellation rule in accordance with BPNN. The system will cancel about 1/3 neurons and reduce 20%¡ã40% memory size under the rule. However, the recognition rate can still achiever up to 85%. For the output structure of the BPNN, we present a binary-code to supersede the one-to-one model. In addition, we use a new ideal about endpoint detection algorithm for the recoding signals. It can avoid disturbance without complex computations.

Thesis (Ph. D.)--University of Washington, 2001. / Includes bibliographical references (p. 80-95).

Thesis (M.S. in Information Systems and Operations)--Naval Postgraduate School, March 2004. / Thesis advisor(s): Russell Gottfried, Monique P. Fargues. Includes bibliographical references (p. 43-45). Also available online.

Thesis (M.S.)--Mississippi State University. Department of Electrical and Computer Engineering. / Title from title screen. Includes bibliographical references.

Thesis (Ph. D.)--University of Hong Kong, 1988.

published_or_final_version / Electrical and Electronic Engineering / Master / Master of Philosophy

No description available.

For many years research in artificial intelligence followed a symbolic paradigm which required a level of knowledge described in terms of rules. More recently subsymbolic approaches have been adopted as a suitable means for studying many problems. There are many search mechanisms which can be used to manipulate subsymbolic components, and in recent years general search methods based on models of natural evolution have become increasingly popular. This thesis examines a hybrid symbolic/subsymbolic approach and the application of evolutionary algorithms to a problem from each of the fields of shape representation (finding an iterated function system for an arbitrary shape), natural language dialogue (tuning parameters so that a particular behaviour can be achieved) and speech recognition (selecting the penalties used by a dynamic programming algorithm in creating a word lattice). These problems were selected on the basis that each should have a fundamentally different interactions at the subsymbolic level. Results demonstrate that for the experiments conducted the evolutionary algorithms performed well in most cases. However, the type of subsymbolic interaction that may occur influences the relative performance of evolutionary algorithms which emphasise either top-down (evolutionary programming - EP) or bottom-up (genetic algorithm - GA) means of solution discovery. For the shape representation problem EP is seen to perform significantly better than a GA, and reasons for this disparity are discussed. Furthermore, EP appears to offer a powerful means of finding solutions to this problem, and so the background and details of the problem are discussed at length. Some novel constraints on the problem's search space are also presented which could be used in related work. For the dialogue and speech recognition problems a GA and EP produce good results with EP performing slightly better. Results achieved with EP have been used to improve the performance of a speech recognition system.

With the development of multimedia technology, speech recognition technology has increasingly become a hotspot of research in recent years. It has a wide range of applications, which deals with recognizing the identity of the speakers that can be classified into speech identification and speech verification according to decision modes.The main work of this thesis is to study and research the techniques, algorithms of speech recognition, thus to create a feasible system to simulate the speech recognition. The research work and achievements are as following: First: The author has done a lot of investigation in the field of speech recognition with the adequate research and study. There are many algorithms about speech recognition, to sum up, the algorithms can divided into two categories, one of them is the direct speech recognition, which means the method can recognize the words directly, and another prefer the second method that recognition based on the training model. Second: find a useable and reasonable algorithm and make research about this algorithm. Besides, the author has studied algorithms, which are used to extract the word's characteristic parameters based on MFCC(Mel frequency Cepstrum Coefficients) , and training the Characteristic parameters based on the GMM(Gaussian mixture mode) . Third: The author has used the MATLAB software and written a program to implement the speech recognition algorithm and also used the speech process toolbox in this program. Generally speaking, whole system includes the module of the signal process, MFCC characteristic parameter and GMM training. Forth: Simulation and analysis the results. The MATLAB system will read the wav file, play it first, and then calculate the characteristic parameters automatically. All content of the speech signal have been distinguished in the last step. In this paper, the author has recorded speech from different people to test the systems and the simulation results shown that when the testing environment is quiet enough and the speaker is the same person to record for 20 times, the performance of the algorithm is approach to 100% for pair of words in different and same syllable. But the result will be influenced when the testing signal is surrounded with certain noise level. The simulation system won’t work with a good output, when the speaker is not the same one for recording both reference and testing signal.