- Publication Date:
- NSF-PAR ID:
- Journal Name:
- ISCA INTERSPEECH-2021
- Page Range or eLocation-ID:
- 1484 to 1488
- Sponsoring Org:
- National Science Foundation
More Like this
Real-time Speaker counting in a cocktail party scenario using Attention-guided Convolutional Neural NetworkMost current speech technology systems are designed to operate well even in the presence of multiple active speakers. However, most solutions assume that the number of co-current speakers is known. Unfortunately, this information might not always be available in real-world applications. In this study, we propose a real-time, single-channel attention-guided Convolutional Neural Network (CNN) to estimate the number of active speakers in overlapping speech. The proposed system extracts higher-level information from the speech spectral content using a CNN model. Next, the attention mechanism summarizes the extracted information into a compact feature vector without losing critical information. Finally, the active speakers are classified using a fully connected network. Experiments on simulated overlapping speech using WSJ corpus show that the attention solution is shown to improve the performance by almost 3% absolute over conventional temporal average pooling. The proposed Attention-guided CNN achieves 76.15% for both Weighted Accuracy and average Recall, and 75.80% Precision on speech segments as short as 20 frames (i.e., 200 ms). All the classification metrics exceed 92% for the attention-guided model in offline scenarios where the input signal is more than 100 frames long (i.e., 1s).
Obeid, Iyad Selesnick (Ed.)Electroencephalography (EEG) is a popular clinical monitoring tool used for diagnosing brain-related disorders such as epilepsy . As monitoring EEGs in a critical-care setting is an expensive and tedious task, there is a great interest in developing real-time EEG monitoring tools to improve patient care quality and efficiency . However, clinicians require automatic seizure detection tools that provide decisions with at least 75% sensitivity and less than 1 false alarm (FA) per 24 hours . Some commercial tools recently claim to reach such performance levels, including the Olympic Brainz Monitor  and Persyst 14 . In this abstract, we describe our efforts to transform a high-performance offline seizure detection system  into a low latency real-time or online seizure detection system. An overview of the system is shown in Figure 1. The main difference between an online versus offline system is that an online system should always be causal and has minimum latency which is often defined by domain experts. The offline system, shown in Figure 2, uses two phases of deep learning models with postprocessing . The channel-based long short term memory (LSTM) model (Phase 1 or P1) processes linear frequency cepstral coefficients (LFCC)  features from each EEGmore »
This paper presents a novel zero-shot learning approach towards personalized speech enhancement through the use of a sparsely active ensemble model. Optimizing speech denoising systems towards a particular test-time speaker can improve performance and reduce run-time complexity. However, test-time model adaptation may be challenging if collecting data from the test-time speaker is not possible. To this end, we propose using an ensemble model wherein each specialist module denoises noisy utterances from a distinct partition of training set speakers. The gating module inexpensively estimates test-time speaker characteristics in the form of an embedding vector and selects the most appropriate specialist module for denoising the test signal. Grouping the training set speakers into non-overlapping semantically similar groups is non-trivial and ill-defined. To do this, we first train a Siamese network using noisy speech pairs to maximize or minimize the similarity of its output vectors depending on whether the utterances derive from the same speaker or not. Next, we perform k-means clustering on the latent space formed by the averaged embedding vectors per training set speaker. In this way, we designate speaker groups and train specialist modules optimized around partitions of the complete training set. Our experiments show that ensemble models made upmore »
Abstract Objective.High-density electromyography (HD-EMG) decomposition algorithms are used to identify individual motor unit (MU) spike trains, which collectively constitute the neural code of movements, to predict motor intent. This approach has advanced from offline to online decomposition, from isometric to dynamic contractions, leading to a wide range of neural-machine interface applications. However, current online methods need offline retraining when applied to the same muscle on a different day or to a different person, which limits their applications in a real-time neural-machine interface. We proposed a deep convolutional neural network (CNN) framework for neural drive estimation, which takes in frames of HD-EMG signals as input, extracts general spatiotemporal properties of MU action potentials, and outputs the number of spikes in each frame. The deep CNN can generalize its application without retraining to HD-EMG data recorded in separate sessions, muscles, or participants. Approach.We recorded HD-EMG signals from the vastus medialis and vastus lateralis muscles from five participants while they performed isometric contractions during two sessions separated by ∼20 months. We identified MU spike trains from HD-EMG signals using a convolutive blind source separation (BSS) method, and then used the cumulative spike train (CST) of these MUs and the HD-EMG signals to train andmore »
In this paper, a novel method using 3D Convolutional Neural Network (3D-CNN) architecture has been proposed for speaker verification in the text-independent setting. One of the main challenges is the creation of the speaker models. Most of the previously-reported approaches create speaker models based on averaging the extracted features from utterances of the speaker, which is known as the d-vector system. In our paper, we propose an adaptive feature learning by utilizing the 3D-CNN s for direct speaker model creation in which, for both development and enrollment phases, an identical number of spoken utterances per speaker is fed to the network for representing the speakers' utterances and creation of the speaker model. This leads to simultaneously capturing the speaker-related information and building a more robust system to cope with within-speaker variation. We demonstrate that the proposed method significantly outperforms the traditional d-vector verification system. Moreover, the proposed system can also be an alternative to the traditional d-vector system which is a one-shot speaker modeling system by utilizing 3D-CNNs.