Search

The theory of sampling method for measuring the mean parameters of quasi-periodic signal is analyzed, and the condition without introducing measurement errors is given. The reason of measuring error caused by non-synchronous sampling and the mechanism of window weighted method to suppress measurement error are analyzed. In particular, the analytical formulas of rectangular convolution window and measurement errors are given. Aiming at the DC component in the input signal of the practical analog-to-digital conversion (ADC) circuit, a series of formulas for parameter measurement of quasi-periodic signal in engineering implementation are derived.

This paper examines the efficient quantization of LSP parameters for very low bit rate vocoder below 300 bps, a new quantization scheme called variable dimension matrix quantization (VDMQ) is presented. In the VDMQ scheme, the extracted LSP parameters matrix with variable dimension is quantized directly without dimension conversion. Based on the distance measure definition between tow LSP matrices with different dimension, the optimal codeword is deduced. Theoretical analysis and experiment results show that the VDMQ scheme performs better than the segment quantization and matrix quantization scheme at very low bit rate. Also, the codebook storage is almost reduced by 90%. The VDMQ scheme provides a new effective approach for efficient LSP parameters quantization at very low bit rate.

This paper analyzes the distortion of enhanced speech based on auditory perceptual properties. A speech enhancement criteria in the sense of minimum perceptual distortion called audible distortion criteria (ADC) is proposed, the gain function of short-time spectral amplitude is deduced then. Through rough estimation for the short-time power spectra and calculation of ADC gain function, a novel speech enhancement method is given. The proposed method can make a good compromise between reinforcing the noise suppression and reducing the speech distortion. It has lower computation complexity and is simple in implementation.

After analyzing the direct sampling algorithm for measuring a periodic signal's RMS value, a triangular window weighted algorithm is proposed for effectively reducing the synchronous error. Analytic formulas of measured RMS value as a function of the relative frequency deviation and initial sampling phase (or time) is derived. Both theory and numerical simulations show that the measurement error of the RMS value is proportional to the relative frequency deviation using the direct sampling algorithm, while it is proportional to the square of the relative frequency deviation using the proposed algorithm. Furthermore, the proposed algorithm is quite suitable for on-line implementation and the measured value responds rapidly to signal's variation. If the frequency deviation is far less than one, very high accurate measuring results can be achieved even if no synchronous sampling technique is introduced.

On the basis of the actions of a signal upon an ideal oscillator and an oscillator-array, the instantaneous energy density spectrum of a signal is defined. Furthermore, definitions of the instantaneous energy, the power density spectrum, the power, the total energy density spectrum and the total energy of a signal are introduced. Properties of power density spectrum and the relations between the power density spectrum and other physical quantities are presented. To meet the needs of real applications, an improvement method to the power density spectrum has also been mentioned.

Most algorithms for speech enhancement in the spectral domain focus on the acquisition of an estimator of the clean speech parameter, such as spectrum or amplitude. Enhanced speech quality with these methods mainly depends upon the accuracy of the estimator. When the signal-to-noise ratio (SNR) becomes lower, e.g., less than 3 dB, the enhanced speech often shows unsatisfactory quality. Here, we propose a new method for speech enhancement in the spectral domain. It is based on our view that speech characteristics are perceived mainly by part of the spectral components which have higher local (or instantaneous) SNRs. These components have a special importance on speech SNR enhancement and speech understanding. Identifying these components in the noisy spectrum with a decision system and constructing enhanced speech just by the same noisy amplitude and phase, the result is similar to those popular algorithms. All that is needed is a local SNR decision for each spectral component. The threshold for the decision is derived and two popular methods for local SNR (LSNR) estimation are suggested. It shows better performance under lower SNR signals. Though the number of these kind of spectral components varies with different signal SNR, any algorithms which can pick up more than 90% of it is sufficient to match any existing methods. Primary considerations and results are shown.