Myung-Jin Bae

A fast pitch searching algorithm using correlation characteristics in CELP vocoder

The major drawback in the code excited linear prediction (CELP) type vocoders is their large computational requirements. In the present paper a simple method is proposed to reduce the pitch searching time in the pitch filter almost without degradation of quality. Based upon the observational regularity of the correlation function of speech, the searching range can be restricted to the positive side in pitch search. This is done by skipping the negative side with the width which is estimated from the previous positive envelope. In addition to that, the maximum number of available lags can be limited by the threshold, LT, which is set on 58 empirically. So, only the limited numbers of lags are considered in pitch search, which is less than a half of that of the full search method. By using the proposed method in pitch search, its required computations are greatly reduced. Experimental results show 51% time reduction almost without lowering the speech quality in segmental SNR measures.<<ETX>>

A fast vector quantization encoding algorithm using multiple projection axes

Computation of nearest neighbor generally requires a large number of expensive distance calculations. In this paper, we present an algorithm which uses multiple projection axes to accelerate the encoding process of VQ by eliminating the necessity of calculating many distances. Since the proposed algorithm rejects those codewords that are impossible to be the nearest codeword, it produces the same output as a conventional full search algorithm. The simulation results confirm the effectiveness of the proposed algorithm.

On a new predictor for the waveform coding of speech signal by using the dual autocorrelation and the sigma-delta technique

Speech waveforms are highly correlated between the adjacent samples. One way of increasing the correlation in speech signals is to simply integrate the input signal prior to coding. The integrated values can be removed by conventional differentiation at the receiver. This emphasizes the low frequencies of speech signals and increases the correlation between adjacent samples. The above arrangement is called as a sigma-delta technique. In this paper, we propose a new predictor which uses such characteristics of dual autocorrelation and the sigma-delta technique. That is, we integrate input signals prior to coding, and then predict the present integrate sample by using two samples, one past and one next. The proposed predictor has higher mean prediction gain of 8.65 dB than that of the CCITT-Recommendation ADPCM.<<ETX>>

On a New Enhancement of Speech Signal Using Non-uniform Sampling and Post Filter

To enhance speech signal by reducing the redundancy within samples that resulted from uniform sampling method like PCM, non-uniform sampling or non-redundant-sample coding methods can be considered. However, it is well known that when conventional non-uniform sampling methods are applied directly to speech signal, the required amount of data in the computation is comparable to or more than that of uniform sampling method. To overcome this problem, a new non-uniform sampling method is proposed, in which non-uniform sampling is applied to the speech signal after using the low pass filter and the remain signals are compensated by the rectified signals with various harmonics frequencies.

A study on the improvement of speaker recognition system by voiced detection

The effect of silence segment elimination on the improvement of a speaker recognition system was investigated in this study. The speaker recognition system is a method to identify input speech of a speaker by the comparison of previously registered and the test speech. The performance of this system greatly depends on a preprocessing stage. The experiments on the extraction of speech segments from speech wave forms during the preprocessing stage were conducted to analyze the rate of recognition. A new measure was developed and proposed, based on the fact that the slope of a valley at the pitch point in speech wave forms is higher when the normalized AMDF was applied to the stationary and transition regions. Voiced segments were extracted from the speech wave forms. From the voice segments, unvoiced segments were then detected using the autocorrelation ratio for the adjacent samples at the front and rear regions of the voiced segments. The results of this study indicated that this proposed method increased the perception rate by approximately 2% but had little effect on recognition time.

On a time reduction of pitch searching by the regular pulse search technique in the CELP vocoder

Code excited linear prediction (CELP) speech coders exhibit good performance at data rates as low as 4800 bps. The major drawback of CELP type coders is the large amount of computational time. In this paper, we propose a new pitch codebook search method that restricts some range of the pitch searching by using the regular pulse excitation (RPE) searching technique. Applying the proposed method to the 8 kbps CELP vocoder, we can get approximately 48% complexity reduction in the pitch search.

The skipping technique: a simple and fast algorithm to find the pitch in CELP vocoder

A fast pitch search algorithm using skipping technique is an efficient method of reducing the computation complexity of pitch search while maintaining a good speech quality in a CELP vocoder. The skipping technique also improves the computation reduction efficiency of conventional fast pitch search methods such as the delta search method by combining with them. Compared with the traditional full search method, the skipping technique combined with the conventional delta search method can reduce the computation time for a pitch search by over 60% with a quality degradation of only 0.57 dB.

The improvement of mobile phone voice quality by bone-conduction device

This paper proposes a new way of reducing noise and enhancing speech signals of mobile phones by installing bone-conduction speakers in ordinary mobile phones. With this new system, the noise from surrounding environments can be reduced and eventually the quality and clarity of voice signals coming out of mobile phones can be improved. The improved voice quality of mobile phones was confirmed by the experiment which measured frequency responses as well as emotional responses before and after the activation of the proposed system.

A Technique for Preventing Noise Induced Hearing Loss Due to Mobile Phone Use Under Noisy Environment

Human auditory acuity decreases naturally due to aging. But recently cases of impaired hearing at a young age are increasing greatly. The biggest reason for such an increase of population with impaired hearing is popularization of various kinds of portable multimedia appliances. Many studies on impaired hearing due to noises caused by the earphone and headphone are being made, but there are few studies on noise-impaired hearing caused directly by mobile phone communication. Based on a precedent inquiry, this study proposes a technique for preventing noise-impaired hearing applying to an active noise reduction technique onto bone conduction speaker. This technique is a method for reducing noises by antiphase oscillation through bone conduction speaker with ambient noises. If the proposed system is applied, the noise level that is actually introduced to audition decreases by more than 12 dB, and such a decreased amount of sound volume fundamentally prevents the factors of noise-caused hearing difficulty due to mobile phone communication. Sensibility test results showed that adequate communication was possible even in such a situation where communication volume was decreased like this.

On a pitch alteration technique of speech using the asymmetry weighted window

To use the speech as an effective communication medium between man and machine, the synthetic speech must have good quality and various voice colors. Speech synthesis coding is classified into three categories: waveform coding, source coding and hybrid coding. To obtain synthetic speech with high quality, synthesis by waveform coding is desired. However, it is difficult to alter the excitation for various voice colors in waveform coding, because it does not divide the speech into excitation and formant components. Thus it is required to alter the excitation (pitch) in waveform coding for synthesis techniques with high quality and various voice colors. This paper examines the method for both improving and indicating the problem of the PSOLA pitch alteration method. It points out the fact that the spectrum distortion appeared because the Hamming window is not appropriate to the characteristic of the glottal wave shape. Therefore the asymmetric weighted window is proposed in order to improve this defect. The experimental procedure is as follows; first, the speech is segmented by the pitch unit with the asymmetric weighted window, and then the segmented speech is synthesized. The results of an experiment with two male speakers and the two female speakers uttering the test sentences are discussed. According to the experimental results, in the case of using the asymmetric weighted window, synthesized speech of high quality with minimum spectrum distortion can be obtained from waveform coding.