Münevver Köküer

Automatic Detection and Recognition of Tonal Bird Sounds in Noisy Environments

This paper presents a study of automatic detection and recognition of tonal bird sounds in noisy environments. The detection of spectro-temporal regions containing bird tonal vocalisations is based on exploiting the spectral shape to identify sinusoidal components in the short-time spectrum. The detection method provides tonal-based feature representation that is employed for automatic bird recognition. The recognition system uses Gaussian mixture models to model 165 different bird syllables, produced by 95 bird species. Standard models, as well as models compensating for the effect of the noise, are employed. Experiments are performed on bird sound recordings corrupted by White noise and real-world environmental noise. The proposed detection method shows high detection accuracy of bird tonal components. The employed tonal-based features show significant recognition accuracy improvements over the Mel-frequency cepstral coefficients, in both standard and noise-compensated models, and strong robustness to mismatch between the training and testing conditions.

Estimation of Voicing-Character of Speech Spectra Based on Spectral Shape

This letter presents a method for estimation of the voicing-character of speech spectra. It is based on a calculation of a similarity between the shape of the signal short-term magnitude spectra and spectra of the frame-analysis window, which is weighted by the signal magnitude spectra. It is demonstrated that the proposed voicing measure is related to the local SNR of noise-corrupted voiced speech. The performance is evaluated for detection of voiced regions in the spectra of speech corrupted by various types of noise. The experimental results in terms of false-acceptance and false-rejection show errors of less than 5% for speech corrupted by white noise at the local SNR of 10 dB and in terms of recognition accuracy obtained by an ASR system using the voicing information estimated by the proposed method and by the full a priori knowledge about the noise show similar recognition performance

Speech Signal Enhancement Based on MAP Algorithm in the ICA Space

This paper presents a novel maximum a posteriori (MAP) denoising algorithm based on the independent component analysis (ICA). We demonstrate that the employment of individual ICA transformations for signal and noise can provide the best estimate within the linear framework. The signal enhancement problem is categorized based on the distribution of signal and noise being Gaussian or non-Gaussian and the estimation rule is derived for each of the categories. Our theoretical analysis shows that under the assumption of a Gaussian noise the proposed algorithm leads to some well-known enhancement techniques, i.e., Wiener filter and sparse code shrinkage. The analysis of the denoising capability shows that the proposed algorithm is most efficient for non-Gaussian signals corrupted by a non-Gaussian noise. We employed the generalized Gaussian model (GGM) to model the distributions of speech and noise. Experimental evaluation is performed in terms of signal-to-noise ratio (SNR) and spectral distortion measure. Experimental results show that the proposed algorithms achieve significant improvement on the enhancement performance in both Gaussian and non-Gaussian noise.

Detection of sinusoidal signals in noise by probabilistic modelling of the spectral magnitude shape and phase continuity

This paper presents a method for detection of sinusoidal signals corrupted by an additive noise in the short-time Fourier domain. The proposed method is based on probabilistic modelling of the spectral magnitude shape and phase continuity around spectral peaks and can deal with both stationary and non-stationary sinusoidal signals. Experimental results are presented for both sinusoidal signals of a constant frequency and frequency varying continuously over time. The performance is analysed in terms of the false acceptance and false rejection error rates of spectral peak detection and also compared to our previous method. Experimental results demonstrate very high detection accuracy in even very strong noisy conditions.

Incorporating the voicing information into HMM-based automatic speech recognition in noisy environments

In this paper, we propose a model for the incorporation of voicing information into a speech recognition system in noisy environments. The employed voicing information is estimated by a novel method that can provide this information for each filter-bank channel and does not require information about the fundamental frequency. The voicing information is modelled by employing the Bernoulli distribution. The voicing model is obtained for each HMM state and mixture by a Viterbi-style training procedure. The proposed voicing incorporation is evaluated both within a standard model and two other models that had compensated for the noise effect, the missing-feature and the multi-conditional training model. Experiments are first performed on noisy speech data from the Aurora 2 database. Significant performance improvements are achieved when the voicing information is incorporated within the standard model as well as the noise-compensated models. The employment of voicing information is also demonstrated on a phoneme recognition task on the noise-corrupted TIMIT database and considerable improvements are observed.

Bird species recognition using HMM-based unsupervised modelling of individual syllables with incorporated duration modelling

This paper presents an HMM-based automatic system for recognition of bird species from audio field recordings. It includes an improved unsupervised modelling of individual bird syllables and duration modelling. The acoustic signal is decomposed into isolated segments, each segment containing a temporal evolution of a detected sinusodal component. Modelling of bird syllables is performed using Hidden Markov models (HMMs). A set of syllables of bird vocalisations is discovered in an unsupervised manner by employing dynamic time warping and agglomerative hierarchical clustering. A novel iterative maximum likelihood procedure is used to train individual HMMs for syllables of each species. Modelling of the state duration is employed in a post-recognition stage by combining the likelihood of the acoustic and duration modelling. Experiments are performed on over 33 hours of field recordings, containing 30 bird species. Evaluations demonstrate that the use of the proposed un-supervised iterative HMM training procedure and the duration modelling provides in average 45% error rate reduction. The presented system recognises bird species with accuracy of 97.8% using 3 seconds of the detected signal.

Speech enhancement based on Sparse Code Shrinkage employing multiple speech models

This paper presents a single-channel speech enhancement system based on the Sparse Code Shrinkage (SCS) algorithm and employment of multiple speech models. The enhancement system consists of two stages: training and enhancement. In the training stage, the Gaussian mixture modelling (GMM) is employed to cluster speech signals in ICA-based transform domain into several categories, and for each category a super-Gaussian model is estimated that is used during the enhancement stage. In the enhancement stage, the estimate of each signal frame is obtained as a weighted average of estimates obtained by using each speech category model. The weights are calculated according to the probability of each category, given the signal enhanced using the conventional SCS algorithm. During the enhancement, the individual speech category models are further adapted at each signal frame. Experimental evaluations are performed on speech signals from the TIMIT database, corrupted by Gaussian noise and three real-world noises, Subway, Street, and Railway noise, from the NOISEX-92 database. Evaluations are performed in terms of segmental SNR, spectral distortion and PESQ measure. Experimental results show that the proposed multi-model SCS enhancement algorithm significantly outperforms the conventional WF, SCS and multi-model WF algorithms.

Underdetermined DOA estimation via independent component analysis and time-frequency masking

This paper presents an algorithm for the estimation of the direction of arrival (DOA) in underdetermined situations, that is, there is more sources than sensors. The algorithm performs the estimation in an iterative manner, each iteration consists of two-steps: first estimation of the DOA of a dominant source via the Independent Component Analysis and then removal of the detected source from themixture via time-frequencymasking. Experiments, performed using speech signals mixed in real environment when only twomicrophones are used but three and four sources are present, demonstrate that the proposed algorithm can estimate the DOAs more accurately than two previously used underdetermined DOA algorithms.

Separation of Harmonic and Speech Signals using Sinusoidal Modeling

This paper studies the problem of separation of two harmonic-based source signals from a single-channel mixture signal based on employment of a sinusoidal model. The sinusoidal model represents the signal as a sum of sine-waves, whose parameters (i.e., frequencies, amplitudes, and phases) are estimated by a least-square method that minimizes the reconstruction error between the model and the mixture signal. A comprehensive evaluation of the performance of the sinusoidal model for separation of simulated harmonic signals with various fundamental frequencies is presented. Very good performance, in terms of signal-to-distortion ratio, is observed without any a-priori knowledge about F0s of individual signals. The studied model is then demonstrated for separation of a mixture of two speech signals.

Incorporating mask modelling for noise-robust automatic speech recognition

In this paper we investigate an incorporation of mask modelling into an HMM-based ASR system. The mask model is estimated for each HMM state and mixture by using a separate Viterbi-style training procedure and it expresses which regions of the spectrum are expected to be uncorrupted by noise for the HMM state. Experimental evaluation is performed on noisy speech data from the Aurora 2 database. Significant performance improvements are achieved when the mask modelling is incorporated within the standard model and two models that had already compensated for the effect of the noise.