No description available.

No description available.

No description available.

No description available.

Adaptive postfiltering has become a common part of speech coding standards based on the Linear Prediction Analysis-by-Synthesis algorithm to decrease audible coding noise. However, a conventional adaptive postfilter is based on empirical assumptions of masking phenomena, which sometimes makes it hard to balance between noise reduction and speech distortion. / This thesis introduces a novel perceptual postfiltering system for low bit rate speech coders. The proposed postfilter works at the decoder, as is the case for the conventional adaptive postfilter. Specific human auditory properties are considered in the postfilter design to improve speech quality. A Gaussian Mixture Model based Minimum Mean Squared Error estimation of the perceptual postfilter is performed with the received information at the decoder. Perceptual postfiltering is then applied to the reconstructed speech to improve speech quality. Test results show that the proposed system gives better perceptual speech quality over conventional adaptive postfiltering.

This thesis presents methods for supervised and unsupervised speaker adaptation of Gaussian mixture speaker models in text-independent speaker verification. The proposed methods are based on an approach which is able to separate speaker and channel variability so that progressive updating of speaker models can be performed while minimizing the influence of the channel variability associated with the adaptation recordings. This approach relies on a joint factor analysis model of intrinsic speaker variability and session variability where inter-session variation is assumed to result primarily from the effects of the transmission channel. These adaptation methods have been evaluated under the adaptation paradigm defined under the NIST 2005 speaker recognition evaluation plan which is based on conversational telephone speech.

No description available.

Thesis (MEng (Research)) -- University of South Australia, 1993

Thesis (PhD) -- University of South Australia, 1999

In most speech communication systems, the presence of background noise causes the quality and intelligibility of speech to degrade, especially when the Signal-to-Noise Ratio (SNR) is low. Numerous speech enhancement techniques have been employed successfully in many applications. However, at low signal-to-noise ratios most of these speech enhancement techniques tend to introduce a perceptually annoying residual noise known as "musical noise". The research presented in this thesis aims to minimize this musical noise and maximize the noise reduction ability of speech enhancement algorithms to improve speech quality in low SNR environments. This thesis proposes two novel speech enhancement algorithms based on Weiner and Kalman filters, and exploit the masking properties of the human auditory system to reduce background noise. The perceptual Wiener filter method uses either temporal or simultaneous masking to adjust the Wiener gain in order to suppress noise below the masking thresholds. The second algorithm involves reshaping the corrupted signal according to the masking threshold in each critical band, followed by Kalman filtering. A comparison of the results from these proposed techniques with those obtained from traditional methods suggests that the proposed algorithms address the problem of noise reduction effectively while decreasing the level of the musical noise. In this thesis, many other existing competitive noise suppression methods have also been discussed and their performance evaluated under different types of noise environments. The performances were evaluated and compared to each other using both objective PESQ measures (ITU-T P.862) and subjective listening tests (ITU-T P.835). The proposed speech enhancement schemes based on the auditory masking model outperformed the other methods that were tested.