An Evaluation of Unsupervised Acoustic Model Training for a Dysarthric Speech Interface

in Proceedings of the 15th Annual Conference of the International Speech Communication Association (INTERSPEECH 2014) (2014, September)

In this paper, we investigate unsupervised acoustic model training approaches for dysarthric-speech recognition. These models are first, frame-based Gaussian posteriorgrams, obtained from Vector ... [more ▼]

In this paper, we investigate unsupervised acoustic model training approaches for dysarthric-speech recognition. These models are first, frame-based Gaussian posteriorgrams, obtained from Vector Quantization (VQ), second, so-called Acoustic Unit Descriptors (AUDs), which are hidden Markov models of phone-like units, that are trained in an unsupervised fashion, and, third, posteriorgrams computed on the AUDs. Experiments were carried out on a database collected from a home automation task and containing nine speakers, of which seven are considered to utter dysarthric speech. All unsupervised modeling approaches delivered significantly better recognition rates than a speaker-independent phoneme recognition baseline, showing the suitability of unsupervised acoustic model training for dysarthric speech. While the AUD models led to the most compact representation of an utterance for the subsequent semantic inference stage, posteriorgram-based representations resulted in higher recognition rates, with the Gaussian posteriorgram achieving the highest slot filling F-score of 97.02%. [less ▲]

Design of nonlinear predictors for adaptive predictive coding of speech signals

in Proceedings of the 21st Telecommunications Forum Telfor (TELFOR) (2013, November)

Linear predictive coding is probably the most frequently used technique in speech signal processing. Its main advantage comes from the analogy of the simplified vocal tract model with speech production ... [more ▼]

Linear predictive coding is probably the most frequently used technique in speech signal processing. Its main advantage comes from the analogy of the simplified vocal tract model with speech production system. However, this neglects nonlinearities in the speech production process. The paper deals with nonlinear prediction of speech based on truncated Volterra series. Long-term one-tap Volterra predictor is designed in order to decrease computational complexity. Further improvements are obtained using frame/subframe structure and fractional delay. [less ▲]

Low-Order Volterra Long-Term Predictors

in Proceedings of the 10. ITG Symposium on Speech Communication (2012, September)

Models based on linear prediction have been used for several decades in different areas of speech signal processing. While the linear approach has led to great advances in the last 40 years, it neglects ... [more ▼]

Models based on linear prediction have been used for several decades in different areas of speech signal processing. While the linear approach has led to great advances in the last 40 years, it neglects nonlinearities present in the speech production mechanism. This paper compares the results of long-term nonlinear prediction based on second-order and third-order Volterra filters. Additional improvement can be obtained using fractionaldelay long-term prediction. Experimental results reveal that the proposed method outperforms linear long-term prediction techniques in terms of prediction gain. [less ▲]

Nonlinear long-term prediction of speech based on truncated Volterra series

in IEEE Transactions on Audio, Speech and Language Processing (2012), 20(3), 1069-1073

Previous studies of nonlinear prediction of speech have been mostly focused on short-term prediction. This paper presents long-term nonlinear prediction based on second-order Volterra filters. It will be ... [more ▼]

Previous studies of nonlinear prediction of speech have been mostly focused on short-term prediction. This paper presents long-term nonlinear prediction based on second-order Volterra filters. It will be shown that the presented predictor can outperform conventional linear prediction techniques in terms of prediction gain and “whiter” residuals. [less ▲]

Switched adaptive quantiser for speech compression based on optimal companding and correlation

in IET Signal Processing (2011), 5(7), 701-707

This study describes a novel adaptive quantiser based on the optimal companding technique. Adaptation is achieved by adjusting the input of the fixed or non-adaptive quantiser according to the estimated ... [more ▼]

This study describes a novel adaptive quantiser based on the optimal companding technique. Adaptation is achieved by adjusting the input of the fixed or non-adaptive quantiser according to the estimated and quantised gain on each particular frame. In such a way better quantiser adaptation to the varying input statistics is provided. Selection of the appropriate bit rate is performed depending on the value of the correlation coefficient ρ on each frame. The decision thresholds for ρ are determined under the condition that the signal to quantisation noise ratio does not drop under 34.3ρdB, satisfying the G.712 standard quality of speech, while decreasing the bit rate. The information about the gain and about the chosen bit rate is then transferred as a side information to a decoder. Although this slightly increases the side information, the overall savings in the bit rate have shown to be substantial. Theoretical and experimental results are provided, which point out the benefits that can be achieved using the proposed algorithm. [less ▲]

DPCM with Forward Gain-Adaptive Quantizer and Simple Switched Predictor for High Quality Speech Signals

in Advances in Electrical and Computer Engineering (2010), 10(4), 95-98

In this article DPCM (Differential Pulse Code Modulation) speech coding scheme with a simple switched first order predictor is presented. Adaptation of the quantizer to the signal variance is performed ... [more ▼]

In this article DPCM (Differential Pulse Code Modulation) speech coding scheme with a simple switched first order predictor is presented. Adaptation of the quantizer to the signal variance is performed for each particular frame. Each frame is classified as high or low correlated, based on the value of the correlation coefficient, then the selection of the appropriate predictor coefficient and bitrate is performed. Low correlated frames are encoded with a higher bitrate, while high correlated frames are encoded with a lower bitrate without the objectionable loss in quality. Theoretical model and experimental results are provided for the proposed algorithm. [less ▲]

Shadows on the Walls: Geometric Interpretation of Fractional Integration

in Journal of Online Mathematics and its Applications (2007), 7

In 2001/2002, Podlubny suggested a solution to the more than 300-years old problem of geometric interpretation of fractional integration (i.e., integration of an arbitrary real order). His geometric ... [more ▼]

In 2001/2002, Podlubny suggested a solution to the more than 300-years old problem of geometric interpretation of fractional integration (i.e., integration of an arbitrary real order). His geometric interpretation for left-sided and right-sided Riemann-Liouville fractional integrals, and for Riesz potential is given in terms of changing time scale with constant order of integration, and also in a case of varying order of integration with constant time parameter. In this article we present animations of such interpretation. [less ▲]