In order to achieve better perceptual coding quality while using fewer bits, a novel perceptual video coding method based on the just-noticeable-distortion （JND） model and the auto-regressive （AR） model is explored. First, a new texture segmentation method exploiting the JND profile is devised to detect and classify texture regions in video scenes. In this step, a spatial-temporal JND model is proposed and the JND energy of every micro-block unit is computed and compared with the threshold. Secondly, in order to effectively remove temporal redundancies while preserving high visual quality, an AR model is applied to synthesize the texture regions. All the parameters of the AR model are obtained by the least-squares method and each pixel in the texture region is generated as a linear combination of pixels taken from the closest forward and backward reference frames. Finally, the proposed method is compared with the H.264/AVC video coding system to demonstrate the performance. Various sequences with different types of texture regions are used in the experiment and the results show that the proposed method can reduce the bit-rate by 15% to 58% while maintaining good perceptual quality.
In order to improve the performance of voice conversion, the fundamental frequency （F0） transformation methods are investigated, and an efficient F0 transformation algorithm is proposed. First, unlike the traditional linear transformation methods, the relationships between F0s and spectral parameters are explored. In each component of the Gaussian mixture model （GMM）, the F0s are predicted from the converted spectral parameters using the support vector regression （SVR） method. Then, in order to reduce the over- smoothing caused by the statistical average of the GMM, a mixed transformation method combining SVR with the traditional mean-variance linear （MVL） conversion is presented. Meanwhile, the adaptive median filter, prevalent in image processing, is adopted to solve the discontinuity problem caused by the frame-wise transformation. Objective and subjective experiments are carried out to evaluate the performance of the proposed method. The results demonstrate that the proposed method outperforms the traditional F0 transformation methods in terms of the similarity and the quality.
A novel emotional speaker recognition system （ESRS） is proposed to compensate for emotion variability. First, the emotion recognition is adopted as a pre-processing part to classify the neutral and emotional speech. Then, the recognized emotion speech is adjusted by prosody modification. Different methods including Gaussian normalization, the Gaussian mixture model （GMM） and support vector regression （SVR） are adopted to define the mapping rules of F0s between emotional and neutral speech, and the average linear ratio is used for the duration modification. Finally, the modified emotional speech is employed for the speaker recognition. The experimental results show that the proposed ESRS can significantly improve the performance of emotional speaker recognition, and the identification rate （IR） is higher than that of the traditional recognition system. The emotional speech with F0 and duration modifications is closer to the neutral one.
A machine learning based speech enhancement method is proposed to improve the intelligibility of whispered speech. A binary mask estimated by a two-class support vector machine （SVM） classifier is used to synthesize the enhanced whisper. A novel noise robust feature called Gammatone feature cosine coefficients （GFCCs） extracted by an auditory periphery model is derived and used for the binary mask estimation. The intelligibility performance of the proposed method is evaluated and compared with the traditional speech enhancement methods. Objective and subjective evaluation results indicate that the proposed method can effectively improve the intelligibility of whispered speech which is contaminated by noise. Compared with the power subtract algorithm and the log-MMSE algorithm, both of which do not improve the intelligibility in lower signal-to-noise ratio （SNR） environments, the proposed method has good performance in improving the intelligibility of noisy whisper. Additionally, the intelligibility of the enhanced whispered speech using the proposed method also outperforms that of the corresponding unprocessed noisy whispered speech.
