Search

© 2004 EUSIPCO. TIFROM [1, 2] is a two channel separation technique, which is well suited to separating audio signals, and in particular, dependent signals that fall outside the scope of conventional BSS applications [1]. One problem with TIFROM however, is degraded performance due to inconsistent estimation of the mixing system. To reduce these inconsistencies, we present a modified algorithm that incorporates k-means clustering [3] and normalised variance, improving upon TIFROM estimation results significantly. To improve TIFROM data efficiency we also include a weighting (running average) function for mixing column estimates. This transforms our modified algorithm into a block based adaptive algorithm with the ability to track a slowly time-varying mixture.

Designing learning experiences supported by information and communication technology (ICT) is becoming an important skill for all academics in the higher education sector. With a range of "quality" measures being implemented and foreshadowed by government, including "dollars" linked to student learning outcomes, all academics will be increasingly asked to examine their instructional strategies and to offer high quality learning opportunities. Sharing learning resources is seen as one strategy to help academics in this change process. As such, online repositories of learning objects are flourishing to encourage the concept of reuse. However, what is lacking are tools to support academics in designing high quality learning environments that incorporate learning objects. This paper presents a prototype tool that uses the concept of a "learning design" as the framework to assist academics in the design process and demonstrates how learning objects can be incorporated.

This work investigates the validity and accuracy of using spatialcues with Time-Delay Estimation (TDE) as a method of segmenting multichannel recorded speech by speaker location. In environments such as meetings where speakers do not significantly alter position, segmentation by speaker location essentially leads to segmentation by speaker ‘ turn’. The proposed system calculates location information using TDEs and spatial cues extracted from multichannel meeting audio recordings. This location information is then input into a simple segmentation algorithm. Experiments have been performed on both theoretical and real meeting recordings with non-overlapping speakers, and theoretical recordings with overlapping speakers. Segmentation results reveal the most robust cue to be a combination of spatial information and TDEs. This cue combination leads to greater segmentation accuracy for classifying individual speakers and detecting overlapping sections than using spatial cues or time-delay information alone.

This paper examines a scalable method for coding the LP residual. The scalable method is capable of increasing the accuracy of the reconstructed speech from a parametric representation at low rates to a more accurate waveform matched representation at higher rates. The method entails pitch length segmentation, decomposition into pulsed and noise components and modeling of the pulsed components using a fixed shape pulse model in a closed-loop, Analysis by Synthesis system. Subjective testing is presented that indicates that in addition to the AbyS modeling, the pulse parameter evolution must be constrained in synthesis. Results indicate that this proposed method is capable of producing perceptually scalable speech quality as the bit rate is increased through 4 kbps.

Decomposition of speech signals into periodic and noise components is widely used in speech coding to facilitate efficient compression. Existing decomposition schemes are too inflexible to model transient changes in the speech signal, require high delay or produce a large parameter set that is not scalable to low rates. This paper proposes a technique that requires only a single frame of speech and produces a scalable decomposition. The latter allows reconstruction accuracy to be varied according to the bit rate available.

This paper reports a scalable method for coding the LP residual. The scalable method is capable of increasing the accuracy of the reconstructed speech from a parametric representation at low rates to a more accurate waveform matched representation at higher rates. The method entails pitch length segmentation, decomposition into pulsed and noise components and modeling of the pulsed components using a fixed shape pulse model in a closed-loop, analysis by synthesis system.

This paper considers low-rate waveform interpolation (WI) coding. It compares the existing, common slowly evolving waveform (SEW) quantisation scheme with two new schemes for representing and quantising the SEW. The first scheme uses a minimum phase estimate to reconstruct the SEW whilst the second scheme uses a pulse model whose parameters are implicitly transmitted in the quantised rapidly evolving waveform (REW). These new schemes maintain or reduce the bit rate required for transmission of the SEW. Results indicate that, for low rate WI coding, necessarily coarse SEW magnitude spectrum quantisation limits the contribution of the SEW to perceptual quality. Perceptual tests indicate that avoiding coarse spectral shape quantisation and using a fixed shape model that lends itself to smooth interpolation, maintains the perceptual quality of the synthesized speech. The proposed fixed shape model requires no bits for transmission, allowing a 12 percent reduction in the overall coder bit rate.

Reducing the bit rate of waveform interpolation speech coders while maintaining the perceptual quality has been the focus of a great deal of research. This letter proposes a new method of slowly evolving waveform (SEW) quantization specifically targeted at low rate coding. The proposed method uses a pulse model whose parameters are implicitly contained in the quantized rapidly evolving waveform (REW) parameters, thus requiring no bits for transmission. Results indicate no degradation in perceptual speech quality when compared to that of the existing SEW quantization method. This retention of perceptual quality is in spite of a 12% reduction in the overall coder bit rate.

In very low rate coding the aim is to accurately represent speech characteristics as efficiently as possible. High coding gains for the spectral features can be achieved through the use of temporal decomposition. Waveform interpolation coders accurately represent the excitation using characteristic waveforms (CWs) extracted at a constant rate. In this paper, the two approaches are combined into a very low rate coder operating at around 1 kbps. It is shown that the evolution of the excitation is related to the evolution of the speech spectrum. To minimise bit rates, the transmission of CWs is adapted to the spectral parameter evolution using the parameters derived from temporal decomposition of the spectral parameters.

Whilst linear prediction is the cornerstone of most modern speech coders, few of these coders incorporate the perceptual characteristics of hearing into the calculation of the linear predictor coefficients (LPCs). This paper proposes a method of incorporating simultaneous masking into the calculation of the LPCs. This modification requires only a modest increase in computational complexity and results in the linear predictor removing more perceptually important information from the input speech signal. This results in a filter that better models the formants of the input speech spectrum. The net effect is that an improvement in quality is achieved for a given bit rate or alternately a bit rate reduction can be achieved while maintaining perceived quality. These results have been confirmed through subjective listening tests.

This paper investigates the properties of analytic transformation of speech into envelope and phase functions. The envelope is shown to evolve slowly with the pitch of the input speech, whilst the phase consists of two components; one evolving slowly with pitch and another that exhibits a more rapid evolution. We investigate decomposing the phase component further using two distinct methods: (a) filtering of the phase in the pitch evolutionary direction and (b) performing a second analytic decomposition of the phase into secondary envelope and phase components. To examine the characteristics of the pitch cycle evolution, the analytic transform is employed in a waveform interpolation (WI) coding structure. The two phase decompositions are then analysed with particular emphasis on quantisation sensitivity and the required transmission rate. Results indicate that the analytic decomposition may offer a degree of scalability to speech coders, especially when employed in coders that exploit pitch evolution such as WI.