Electrical Engineering And Systems Science

Speech enhancement in the time domain is becoming increasingly popular in recent years, due to its capability to jointly enhance both the magnitude and the phase of speech. In this work, we propose a dense convolutional network (DCN) with self-attention for speech enhancement in the time domain. DCN is an encoder and decoder based architecture with skip connections. Each layer in the encoder and the decoder comprises a dense block and an attention module. Dense blocks and attention modules help in feature extraction using a combination of feature reuse, increased network depth, and maximum context aggregation. Furthermore, we reveal previously unknown problems with a loss based on the spectral magnitude of enhanced speech. To alleviate these problems, we propose a novel loss based on magnitudes of enhanced speech and a predicted noise. Even though the proposed loss is based on magnitudes only, a constraint imposed by noise prediction ensures that the loss enhances both magnitude and phase. Experimental results demonstrate that DCN trained with the proposed loss substantially outperforms other state-of-the-art approaches to causal and non-causal speech enhancement.

Recent approaches for music source separation are almost exclusively based on deep neural networks, mostly employing recurrent neural networks (RNNs). Although RNNs are in many cases superior than other types of deep neural networks for sequence processing, they are known to have specific difficulties in training and parallelization, especially for the typically long sequences encountered in music source separation. In this paper we present a use-case of replacing RNNs with depth-wise separable (DWS) convolutions, which are a lightweight and faster variant of the typical convolutions. We focus on singing voice separation, employing an RNN architecture, and we replace the RNNs with DWS convolutions (DWS-CNNs). We conduct an ablation study and examine the effect of the number of channels and layers of DWS-CNNs on the source separation performance, by utilizing the standard metrics of signal-to-artifacts, signal-to-interference, and signal-to-distortion ratio. Our results show that by replacing RNNs with DWS-CNNs yields an improvement of 1.20, 0.06, 0.37 dB, respectively, while using only 20.57% of the amount of parameters of the RNN architecture.

The purpose of speech dereverberation is to remove quality-degrading effects of a time-invariant impulse response filter from the signal. In this report, we describe an approach to speech dereverberation that involves joint estimation of the dry speech signal and of the room impulse response. We explore deep learning models that apply to each task separately, and how these can be combined in a joint model with shared parameters.

Neural speaker embeddings trained using classification objectives have demonstrated state-of-the-art performance in multiple applications. Typically, such embeddings are trained on an out-of-domain corpus on a single task e.g., speaker classification, albeit with a large number of classes (speakers). In this work, we reformulate embedding training under the meta-learning paradigm. We redistribute the training corpus as an ensemble of multiple related speaker classification tasks, and learn a representation that generalizes better to unseen speakers. First, we develop an open source toolkit to train x-vectors that is matched in performance with pre-trained Kaldi models for speaker diarization and speaker verification applications. We find that different bottleneck layers in the architecture variedly favor different applications. Next, we use two meta-learning strategies, namely prototypical networks and relation networks, to improve over the x-vector embeddings. Our best performing model achieves a relative improvement of 12.37% and 7.11% in speaker error on the DIHARD II development corpus and the AMI meeting corpus, respectively. We analyze improvements across different domains in the DIHARD corpus. Notably, on the challenging child speech domain, we study the relation between child age and the diarization performance. Further, we show reductions in equal error rate for speaker verification on the SITW corpus (7.68%) and the VOiCES challenge corpus (8.78%). We observe that meta-learning particularly offers benefits in challenging acoustic conditions and recording setups encountered in these corpora. Our experiments illustrate the applicability of meta-learning as a generalized learning paradigm for training deep neural speaker embeddings.

In this paper, we propose to address the issue of the lack of strongly labeled data by using pseudo strongly labeled data that is approximated using Convolutive Nonnegative Matrix Factorization (CNMF). Using this pseudo strongly labeled data, we then train a new architecture combining Convolutional Neural Network (CNN) with Macaron Net (MN), which we term it as Convolutional Macaron Net (CMN). As opposed to the Mean-Teacher approach which trains two similar models synchronously, we propose to train two different CMNs synchronously where one of the models will provide the frame-level prediction while the other will provide the clip level prediction. Based on our proposed framework, our system outperforms the baseline system of the Detection and Classification of Acoustic Scenes and Events (DCASE) 2020 Challenge Task 4 by a margin of over 10%. By comparing with the first place of the challenge which utilize a combination of CNN and Conformer, our system also marginally wins it by 0.3%.

Research has shown that trust is an essential aspect of human-computer interaction directly determining the degree to which the person is willing to use the system. An automatic prediction of the level of trust that a user has on a certain system could be used to attempt to correct potential distrust by having the system take relevant actions like, for example, explaining its actions more thoroughly. In this work, we explore the feasibility of automatically detecting the level of trust that a user has on a virtual assistant (VA) based on their speech. We use a dataset collected for this purpose, containing human-computer speech interactions where subjects were asked to answer various factual questions with the help of a virtual assistant, which they were led to believe was either very reliable or unreliable. We find that the subject's speech can be used to detect which type of VA they were using, which could be considered a proxy for the user's trust toward the VA's abilities, with an accuracy up to 76\%, compared to a random baseline of 50\%. These results are obtained using features that have been previously found useful for detecting speech directed to infants and non-native speakers.

In this paper, we envision a web-based framework that can help anyone, anywhere around the world record a short speech task, and analyze the recorded data to screen for Parkinson's disease (PD). We collected data from 726 unique participants (262 PD, 38% female; 464 non-PD, 65% female; average age: 61) -- from all over the US and beyond. A small portion of the data was collected in a lab setting to compare quality. The participants were instructed to utter a popular pangram containing all the letters in the English alphabet "the quick brown fox jumps over the lazy dog..". We extracted both standard acoustic features (Mel Frequency Cepstral Coefficients (MFCC), jitter and shimmer variants) and deep learning based features from the speech data. Using these features, we trained several machine learning algorithms. We achieved 0.75 AUC (Area Under The Curve) performance on determining presence of self-reported Parkinson's disease by modeling the standard acoustic features through the XGBoost -- a gradient-boosted decision tree model. Further analysis reveal that the widely used MFCC features and a subset of previously validated dysphonia features designed for detecting Parkinson's from verbal phonation task (pronouncing 'ahh') contains the most distinct information. Our model performed equally well on data collected in controlled lab environment as well as 'in the wild' across different gender and age groups. Using this tool, we can collect data from almost anyone anywhere with a video/audio enabled device, contributing to equity and access in neurological care.

Oral cancer speech is a disease which impacts more than half a million people worldwide every year. Analysis of oral cancer speech has so far focused on read speech. In this paper, we 1) present and 2) analyse a three-hour long spontaneous oral cancer speech dataset collected from YouTube. 3) We set baselines for an oral cancer speech detection task on this dataset. The analysis of these explainable machine learning baselines shows that sibilants and stop consonants are the most important indicators for spontaneous oral cancer speech detection.

Because of its streaming nature, recurrent neural network transducer (RNN-T) is a very promising end-to-end (E2E) model that may replace the popular hybrid model for automatic speech recognition. In this paper, we describe our recent development of RNN-T models with reduced GPU memory consumption during training, better initialization strategy, and advanced encoder modeling with future lookahead. When trained with Microsoft's 65 thousand hours of anonymized training data, the developed RNN-T model surpasses a very well trained hybrid model with both better recognition accuracy and lower latency. We further study how to customize RNN-T models to a new domain, which is important for deploying E2E models to practical scenarios. By comparing several methods leveraging text-only data in the new domain, we found that updating RNN-T's prediction and joint networks using text-to-speech generated from domain-specific text is the most effective.

In this technical report, we present a joint effort of four groups, namely GT, USTC, Tencent, and UKE, to tackle Task 1 - Acoustic Scene Classification (ASC) in the DCASE 2020 Challenge. Task 1 comprises two different sub-tasks: (i) Task 1a focuses on ASC of audio signals recorded with multiple (real and simulated) devices into ten different fine-grained classes, and (ii) Task 1b concerns with classification of data into three higher-level classes using low-complexity solutions. For Task 1a, we propose a novel two-stage ASC system leveraging upon ad-hoc score combination of two convolutional neural networks (CNNs), classifying the acoustic input according to three classes, and then ten classes, respectively. Four different CNN-based architectures are explored to implement the two-stage classifiers, and several data augmentation techniques are also investigated. For Task 1b, we leverage upon a quantization method to reduce the complexity of two of our top-accuracy three-classes CNN-based architectures. On Task 1a development data set, an ASC accuracy of 76.9\% is attained using our best single classifier and data augmentation. An accuracy of 81.9\% is then attained by a final model fusion of our two-stage ASC classifiers. On Task 1b development data set, we achieve an accuracy of 96.7\% with a model size smaller than 500KB. Code is available: this https URL.