Seongkyu Mun

Deep Neural Network based learning and transferring mid-level audio features for acoustic scene classification

Deep Neural Network (DNN) based transfer learning has been shown to be effective in Visual Object Classification (VOC) for complementing the deficit of target domain training samples by adapting classifiers that have been pre-trained for other large-scaled DataBase (DB). Although there exists an abundance of acoustic data, it can also be said that datasets of specific acoustic scenes are sparse for training Acoustic Scene Classification (ASC) models. By exploiting VOC DNNs ability of learning beyond its pre-trained environments, this paper proposes DNN based transfer learning for ASC. Effectiveness of the proposed method is demonstrated on the database of IEEE DCASE Challenge 2016 Task 1 and home surveillance environment via representative experiments. Its improved performance is verified by comparing it to prominent conventional methods.

Deep neural network bottleneck features for acoustic event recognition

Bottleneck features have been shown to be effective in improving the accuracy of speaker recognition, language identification and automatic speech recognition. However, few works have focused on bottleneck features for acoustic event recognition. This paper proposes a novel acoustic event recognition framework using bottleneck features derived from a Deep Neural Network (DNN). In addition to conventional features (MFCC, Mel-spectrum, etc.), this paper employs rhythm, timbre, and spectrum-statistics features for effectively extracting acoustic characteristics from audio signals. The effectiveness of the proposed method is demonstrated on a database of real life recordings via experiments, and its robust performance is verified by comparing to conventional methods.

Maximum likelihood Linear Dimension Reduction of heteroscedastic feature for robust Speaker Recognition

This paper analyzes heteroscedasticity in i-vector for robust forensics and surveillance speaker recognition system. Linear Discriminant Analysis (LDA), a widely-used linear dimension reduction technique, assumes that classes are homoscedastic within a same covariance. In this paper it is assumed that general speech utterances contain both homoscedastic and heteroscedastic elements. We show the validity of this assumption by employing several analyses and also demonstrate that dimension reduction using principal components is feasible. To effectively handle the presence of heteroscedastic and homoscedastic elements, we propose a fusion approach of applying both LDA and Heteroscedastic-LDA (HLDA). The experiments are conducted to show its effectiveness and compare to other methods using the telephone database of National Institute of Standards and Technology (NIST) Speaker Recognition Evaluation (SRE) 2010 extended.

Robust speaker direction estimation with microphone array using NMF for smart TV interaction

This paper proposes a robust speaker direction estimation method based on a microphone array for voice based interaction with smart TV. The proposed method uses speech basis and associated weights of non-negative matrix factorization for finding the speaker independent utterance direction from input signal with noise. The experimental results of the speaker direction estimation in real acoustic environment validate the effectiveness of the proposed algorithm in terms of representative performance measures compared to the conventional methods.

Non-negative matrix factorisation-based subband decomposition for acoustic source localisation

A novel non-negative matrix factorisation (NMF)-based subband decomposition in frequency–spatial domain for acoustic source localisation using a microphone array is introduced. The proposed method decomposes source and noise subband and emphasises source dominant frequency bins for more accurate source representation. By employing NMF, delay basis vectors and their subband information in frequency–spatial domain for each frame is extracted. The proposed algorithm is evaluated in both simulated noise and real noise with a speech corpus database. Experimental results clearly indicate that the algorithm performs more accurately than other conventional algorithms under both reverberant and noisy acoustic environments.