Despite the progress of deep neural networks over the last decade, the state-of-the-art speech recognizers in noisy environment conditions are still far from reaching satisfactory performance. Methods to improve noise robustness usually include adding components to the recognition system that often need optimization. For this reason, data augmentation of the input features derived from the Short-Time Fourier Transform (STFT) has become a popular approach. However, for many speech processing tasks, there is an evidence that the combination of STFT-based and Hilbert–Huang transform (HHT)-based features improves the overall performance. The Hilbert spectrum can be obtained using adaptive mode decomposition (AMD) techniques, which are noise-robust and suitable for non-linear and non-stationary signal analysis. In this study, we developed a DeepSpeech2-based recognition system by adding a combination of STFT and HHT spectrum-based features. We propose several ways to combine those features at different levels of the neural network. All evaluations were performed using the WSJ and CHiME-4 databases. Experimental results show that combining STFT and HHT spectra leads to a 5–7% relative improvement in noisy speech recognition.
Cough is a common symptom associated with respiratory diseases and its analysis plays a crucial role in monitoring the health conditions of affected persons. Traditional cough detection approaches largely fail to identify single cough boundaries when continuous coughs are present, consequently limiting their suitability for effective cough monitoring. In this research, we propose a novel deep learning system for the efficient detection and monitoring of cough events in audio recordings. Our detection pipeline consists of three key steps. First, we perform voice activity detection to eliminate audio silences and focus on relevant segments. Next, we employ a cough classification technique to identify the presence of cough within those audio segments. Finally, we implement cough event detection using a high-performance classification-regression fusion method. Our approach differs from the traditional audio event detection methods in several notable ways: (1) we incorporate a teacher-student framework for the training of our detection model, (2) instead of relying on specific audio features such as MFCC or Mel Spectrogram, our end-to-end system takes the raw audio signal directly as input and outputs the cough boundary timings, (3) the proposed method is general enough to be used for various other sound event monitoring tasks. Our detection model demonstrated strong performance and robustness on both the in-house and public datasets, by achieving cough event detection error-rate scores of 0.31 and 0.32, respectively, which is several times lower than other models. The comparative cough monitoring evaluation of our approach against systems such as the Leicester Cough Monitor and XGBoost demonstrates our method’s superiority by achieving the lowest average hourly symmetric mean absolute error (sMAPE) of 8.48%. The code is available at https://github.com/FengZongyao/Neural-Cough-Counter.
24 October 2019 in IEEE Journal of Selected Topics in Signal Processing
Availability of large amounts of raw unlabeled data has sparked the recent surge in semi-supervised learning research. In most works, however, it is assumed that labeled and unlabeled data come from the same distribution. This restriction is removed in the self-taught learning algorithm where unlabeled data can be different, but nevertheless have similar structure. First, a representation is learned from the unlabeled samples by decomposing their data matrix into two matrices called bases and activations matrix respectively. This procedure is justified by the assumption that each sample is a linear combination of the columns in the bases matrix which can be viewed as high level features representing the knowledge learned from the unlabeled data in an unsupervised way. Next, activations of the labeled data are obtained using the bases which are kept fixed. Finally, a classifier is built using these activations instead of the original labeled data. In this work, we investigated the performance of three popular methods for matrix decomposition: Principal Component Analysis (PCA), Non-negative Matrix Factorization (NMF) and Sparse Coding (SC) as unsupervised high level feature extractors for the self-taught learning algorithm. We implemented this algorithm for the music genre classification task using two different databases: one as unlabeled data pool and the other as data for supervised classifier training. Music pieces come from 10 and 6 genres for each database respectively, while only one genre is common for the both of them. Results from wide variety of experimental settings show that the self-taught learning method improves the classification rate when the amount of labeled data is small and, more interestingly, that consistent improvement can be achieved for a wide range of unlabeled data sizes. The best performance among the matrix decomposition approaches was shown by the Sparse Coding method.