In TV services, dialogue level personalization is key to meeting user preferences and needs. When dialogue and background sounds are not separately available from the production stage, Dialogue Separation (DS) can estimate them to enable personalization. DS was shown to provide clear benefits for the end user. Still, the estimated signals are not perfect, and some leakage can be introduced. This is undesired, especially during passages without dialogue. We propose to combine DS and Voice Activity Detection (VAD), both recently proposed for TV audio. When their combination suggests dialogue inactivity, background components leaking in the dialogue estimate are reassigned to the background estimate. A clear improvement of the audio quality is shown for dialogue-free signals, without performance drops when dialogue is active. A post-processed VAD estimate with improved detection accuracy is also generated. It is concluded that DS and VAD can improve each other and are better used together.
Recent works show that speech separation guided diarization (SSGD) is an increasingly promising direction, mainly thanks to the recent progress in speech separation. It performs diarization by first separating the speakers and then applying voice activity detection (VAD) on each separated stream. In this work we conduct an in-depth study of SSGD in the conversational telephone speech (CTS) domain, focusing mainly on low-latency streaming diarization applications. We consider three state-of-the-art speech separation (SSep) algorithms and study their performance both in online and offline scenarios, considering non-causal and causal implementations as well as continuous SSep (CSS) windowed inference. We compare different SSGD algorithms on two widely used CTS datasets: CALLHOME and Fisher Corpus (Part 1 and 2) and evaluate both separation and diarization performance. To improve performance, a novel, causal and computationally efficient leakage removal algorithm is proposed, which significantly decreases false alarms. We also explore, for the first time, fully end-to-end SSGD integration between SSep and VAD modules. Crucially, this enables fine-tuning on real-world data for which oracle speakers sources are not available. In particular, our best model achieves 8.8% DER on CALLHOME, which outperforms the current state-of-the-art end-to-end neural diarization model, despite being trained on an order of magnitude less data and having significantly lower latency, i.e., 0.1 vs. 1 s. Finally, we also show that the separated signals can be readily used also for automatic speech recognition, reaching performance close to using oracle sources in some configurations.
Whispering is a ubiquitous mode of communication that humans use daily. Despite this, whispered speech has been poorly served by existing speech technology due to a shortage of resources and processing methodology. To remedy this, this paper provides a processing framework that enables access to large and unique data of high-quality whispered speech. We obtain the data from recordings submitted to online platforms as part of the ASMR media-cultural phenomenon. We describe our processing pipeline and a method for improved whispered activity detection (WAD) in the ASMR data. To efficiently obtain labelled, clean whispered speech, we complement the automatic WAD by using Edyson, a bulk audio-annotation tool with human-in-the-loop. We also tackle a problem particular to ASMR: separation of whisper from other acoustic triggers present in the genre. We show that the proposed WAD and the efficient labelling allows to build extensively augmented data and train a classifier that extracts clean whisper segments from ASMR audio. Our large and growing dataset enables whisper-capable, data-driven speech technology and linguistic analysis. It also opens opportunities in e.g. HCI as a resource that may elicit emotional, psychological and neuro-physiological responses in the listener.
This paper introduces the SWANT team entry to the ICASSP 2023 AEC Challenge. We submit a system that cascades a linear filter with a neural post-filter. Particularly, we adopt sub-band processing to handle full-band signals and shape the network with multi-task learning, where dual signal voice activity detection (DSVAD) and echo estimation are adopted as auxiliary tasks. Moreover, we particularly improve the time frequency convolution module (TFCM) to increase the receptive field using small convolution kernels. Finally, our system has ranked 4th in ICASSP 2023 AEC Challenge Non-personalized track.
Since diarization and source separation of meeting data are closely related tasks, we here propose an approach to perform the two objectives jointly. It builds upon the target-speaker voice activity detection (TS-VAD) diarization approach, which assumes that initial speaker embeddings are available. We replace the final combined speaker activity estimation network of TS-VAD with a network that produces speaker activity estimates at a time-frequency resolution. Those act as masks for source extraction, either via masking or via beamforming. The technique can be applied both for single-channel and multi-channel input and, in both cases, achieves a new state-of-the-art word error rate (WER) on the LibriCSS meeting data recognition task. We further compute speaker-aware and speaker-agnostic WERs to isolate the contribution of diarization errors to the overall WER performance.
Transformer-based end-to-end neural speaker diarization (EEND) models utilize the multi-head self-attention (SA) mechanism to enable accurate speaker label prediction in overlapped speech regions. In this study, to enhance the training effectiveness of SA-EEND models, we propose the use of auxiliary losses for the SA heads of the transformer layers. Specifically, we assume that the attention weight matrices of an SA layer are redundant if their patterns are similar to those of the identity matrix. We then explicitly constrain such matrices to exhibit specific speaker activity patterns relevant to voice activity detection or overlapped speech detection tasks. Consequently, we expect the proposed auxiliary losses to guide the transformer layers to exhibit more diverse patterns in the attention weights, thereby reducing the assumed redundancies in the SA heads. The effectiveness of the proposed method is demonstrated using the simulated and CALLHOME datasets for two-speaker diarization tasks, reducing the diarization error rate of the conventional SA-EEND model by 32.58% and 17.11%, respectively.
While much of modern speech and audio processing relies on deep neural networks trained using fixed audio representations, recent studies suggest great potential in acoustic frontends learnt jointly with a backend. In this study, we focus specifically on learnable filterbanks. Prior studies have reported that in frontends using learnable filterbanks initialised to a mel scale, the learned filters do not differ substantially from their initialisation. Using a Gabor-based filterbank, we investigate the sensitivity of a learnable filterbank to its initialisation using several initialisation strategies on two audio tasks: voice activity detection and bird species identification. We use the Jensen-Shannon Distance and analysis of the learned filters before and after training. We show that although performance is overall improved, the filterbanks exhibit strong sensitivity to their initialisation strategy. The limited movement from initialised values suggests that alternate optimisation strategies may allow a learnable frontend to reach better overall performance.
We describe the system used by our team for the VoxCeleb Speaker Recognition Challenge 2022 (VoxSRC 2022) in the speaker diarization track. Our solution was designed around a new combination of voice activity detection algorithms that uses the strengths of several systems. We introduce a novel multi stream approach with a decision protocol based on classifiers entropy. We called this method a multi-stream voice activity detection and used it with standard baseline diarization embeddings, clustering and resegmentation. With this work, we successfully demonstrated that using a strong baseline and working only on voice activity detection, one can achieved close to state-of-theart results.
Sleep behaviour and in-bed movements contain rich information on the neurophysiological health of people, and have a direct link to the general well-being and quality of life. Standard clinical practices rely on polysomnography for sleep assessment; however, it is intrusive, performed in unfamiliar environments and requires trained personnel. Progress has been made on less invasive sensor technologies, such as actigraphy, but clinical validation raises concerns over their reliability and precision. Additionally, the field lacks a widely acceptable algorithm, with proposed approaches ranging from raw signal or feature thresholding to data-hungry classification models, many of which are unfamiliar to medical staff. This paper proposes an online Bayesian probabilistic framework for objective (in)activity detection and segmentation based on clinically meaningful joint kinematics, measured by a custom-made wearable sensor. Intuitive three-dimensional visualisations of kinematic timeseries were accomplished through dimension reduction based preprocessing, offering out-of-the-box framework explainability potentially useful for clinical monitoring and diagnosis. The proposed framework attained up to 99.2\% $F_1$-score and 0.96 Pearson's correlation coefficient in, respectively, the posture change detection and inactivity segmentation tasks. The work paves the way for a reliable home-based analysis of movements during sleep which would serve patient-centred longitudinal care plans.
Recently, grant-free transmission paradigm has been introduced for massive Internet of Things (IoT) networks to save both time and bandwidth and transmit the message with low latency. In order to accurately decode the message of each device at the base station (BS), first, the active devices at each transmission frame must be identified. In this work, first we investigate the problem of activity detection as a threshold comparing problem. We show the convexity of the activity detection method through analyzing its probability of error which makes it possible to find the optimal threshold for minimizing the activity detection error. Consequently, to achieve an optimum solution, we propose a deep learning (DL)-based method called convolutional neural network (CNN)-activity detection (AD). In order to make it more practical, we consider unknown and time-varying activity rate for the IoT devices. Our simulations verify that our proposed CNN-AD method can achieve higher performance compared to the existing non-Bayesian greedy-based methods. This is while existing methods need to know the activity rate of IoT devices, while our method works for unknown and even time-varying activity rates
Emulating the human ability to solve the cocktail party problem, i.e., focus on a source of interest in a complex acoustic scene, is a long standing goal of audio source separation research. Much of this research investigates separating speech from noise, speech from speech, musical instruments from each other, or sound events from each other. In this paper, we focus on the cocktail fork problem, which takes a three-pronged approach to source separation by separating an audio mixture such as a movie soundtrack or podcast into the three broad categories of speech, music, and sound effects (SFX - understood to include ambient noise and natural sound events). We benchmark the performance of several deep learning-based source separation models on this task and evaluate them with respect to simple objective measures such as signal-to-distortion ratio (SDR) as well as objective metrics that better correlate with human perception. Furthermore, we thoroughly evaluate how source separation can influence downstream transcription tasks. First, we investigate the task of activity detection on the three sources as a way to both further improve source separation and perform transcription. We formulate the transcription tasks as speech recognition for speech and audio tagging for music and SFX. We observe that, while the use of source separation estimates improves transcription performance in comparison to the original soundtrack, performance is still sub-optimal due to artifacts introduced by the separation process. Therefore, we thoroughly investigate how remixing of the three separated source stems at various relative levels can reduce artifacts and consequently improve the transcription performance. We find that remixing music and SFX interferences at a target SNR of 17.5 dB reduces speech recognition word error rate, and similar impact from remixing is observed for tagging music and SFX content.
Trajectory-User Linking (TUL) is a relatively new mobility classification task in which anonymous trajectories are linked to the users who generated them. With applications ranging from personalized recommendations to criminal activity detection, TUL has received increasing attention over the past five years. While research has focused mainly on learning deep representations that capture complex spatio-temporal mobility patterns unique to individual users, we demonstrate that visit patterns are highly unique among users and thus simple heuristics applied directly to the raw data are sufficient to solve TUL. More specifically, we demonstrate that a single check-in per trajectory is enough to correctly predict the identity of the user up to 85% of the time. Moreover, by using a non-parametric classifier, we scale up TUL to over 100k users which is an increase over state-of-the-art by three orders of magnitude. Extensive empirical analysis on four real-world datasets (Brightkite, Foursquare, Gowalla and Weeplaces) compares our findings to state-of-the-art results, and more importantly validates our claim that TUL is easier than commonly believed.
Voice Activity Detection (VAD) is a fundamental module in many audio applications. Recent state-of-the-art VAD systems are often based on neural networks, but they require a computational budget that usually exceeds the capabilities of a small battery-operated device when preserving the performance of larger models. In this work, we rely on the input from a bone conduction microphone (BCM) to design an efficient VAD (BC-VAD) robust against residual non-stationary noises originating from the environment or speakers not wearing the BCM.We first show that a larger VAD system (58k parameters) achieves state-of-the-art results on a publicly available benchmark but fails when running on bone conduction signals. We then compare its variant BC-VAD (5k parameters and trained on BC data) with a baseline especially designed for a BCM and show that the proposed method achieves better performances under various metrics while keeping the realtime processing requirement for a microcontroller.
Grant-free random access is an effective technology for enabling low-overhead and low-latency massive access, where joint activity detection and channel estimation (JADCE) is a critical issue. Although existing compressive sensing algorithms can be applied for JADCE, they usually fail to simultaneously harvest the following properties: effective sparsity inducing, fast convergence, robust to different pilot sequences, and adaptive to time-varying networks. To this end, we propose an unfolding framework for JADCE based on the proximal gradient method. Specifically, we formulate the JADCE problem as a group-row-sparse matrix recovery problem and leverage a minimax concave penalty rather than the widely-used $\ell_1$-norm to induce sparsity. We then develop a proximal gradient-based unfolding neural network that parameterizes the algorithmic iterations. To improve convergence rate, we incorporate momentum into the unfolding neural network, and prove the accelerated convergence theoretically. Based on the convergence analysis, we further develop an adaptive-tuning algorithm, which adjusts its parameters to different signal-to-noise ratio settings. Simulations show that the proposed unfolding neural network achieves better recovery performance, convergence rate, and adaptivity than current baselines.
This paper considers joint user identification and channel estimation (JUICE) in grant-free access with a \emph{clustered} user activity pattern. In particular, we address the JUICE in massive machine-type communications (mMTC) network under correlated Rayleigh fading channels with unknown channel covariance matrices. We formulate the JUICE problem as a maximum \emph{a posteriori} probability (MAP) problem with properly chosen priors to incorporate the partial knowledge of the UEs' clustered activity and the unknown covariance matrices. We derive a computationally-efficient algorithm based on alternating direction method of multipliers (ADMM) to solve the MAP problem iteratively via a sequence of closed-form updates. Numerical results highlight the significant improvements brought by the proposed approach in terms of channel estimation and activity detection performances for clustered user activity patterns.
Load event detection is the fundamental step for the event-based non-intrusive load monitoring (NILM). However, existing event detection methods with fixed parameters may fail in coping with the inherent multi-timescale characteristics of events and their event detection accuracy is easily affected by the load fluctuation. In this regard, this paper extends our previously designed two-stage event detection framework, and proposes a novel multi-timescale event detection method based on the principle of minimum description length (MDL). Following the completion of step-like event detection in the first stage, a long-transient event detection scheme with variable-length sliding window is designed for the second stage, which is intended to provide the observation and characterization of the same event at different time scales. In that, the context information in the aggregated load data is mined by motif discovery, and then based on the MDL principle, the proper observation scales are selected for different events and the corresponding detection results are determined. In the post-processing step, a load fluctuation location method based on voice activity detection (VAD) is proposed to identify and remove the unreasonable events caused by fluctuations. Based on newly proposed evaluation metrics, the comparison tests on public and private datasets demonstrate that our method achieves higher detection accuracy and integrity for events of various appliances across different scenarios.
Although the UA-Speech and TORGO databases of control and dysarthric speech are invaluable resources made available to the research community with the objective of developing robust automatic speech recognition systems, they have also been used to validate a considerable number of automatic dysarthric speech classification approaches. Such approaches typically rely on the underlying assumption that recordings from control and dysarthric speakers are collected in the same noiseless environment using the same recording setup. In this paper, we show that this assumption is violated for the UA-Speech and TORGO databases. Using voice activity detection to extract speech and non-speech segments, we show that the majority of state-of-the-art dysarthria classification approaches achieve the same or a considerably better performance when using the non-speech segments of these databases than when using the speech segments. These results demonstrate that such approaches trained and validated on the UA-Speech and TORGO databases are potentially learning characteristics of the recording environment or setup rather than dysarthric speech characteristics. We hope that these results raise awareness in the research community about the importance of the quality of recordings when developing and evaluating automatic dysarthria classification approaches.
We propose Token Turing Machines (TTM), a sequential, autoregressive Transformer model with memory for real-world sequential visual understanding. Our model is inspired by the seminal Neural Turing Machine, and has an external memory consisting of a set of tokens which summarise the previous history (i.e., frames). This memory is efficiently addressed, read and written using a Transformer as the processing unit/controller at each step. The model's memory module ensures that a new observation will only be processed with the contents of the memory (and not the entire history), meaning that it can efficiently process long sequences with a bounded computational cost at each step. We show that TTM outperforms other alternatives, such as other Transformer models designed for long sequences and recurrent neural networks, on two real-world sequential visual understanding tasks: online temporal activity detection from videos and vision-based robot action policy learning. Code is publicly available at: https://github.com/google-research/scenic/tree/main/scenic/projects/token_turing
End-to-end diarization presents an attractive alternative to standard cascaded diarization systems because a single system can handle all aspects of the task at once. Many flavors of end-to-end models have been proposed but all of them require (so far non-existing) large amounts of annotated data for training. The compromise solution consists in generating synthetic data and the recently proposed simulated conversations (SC) have shown remarkable improvements over the original simulated mixtures (SM). In this work, we create SC with multiple speakers per conversation and show that they allow for substantially better performance than SM, also reducing the dependence on a fine-tuning stage. We also create SC with wide-band public audio sources and present an analysis on several evaluation sets. Together with this publication, we release the recipes for generating such data and models trained on public sets as well as the implementation to efficiently handle multiple speakers per conversation and an auxiliary voice activity detection loss.
The increasing number of surveillance cameras and security concerns have made automatic violent activity detection from surveillance footage an active area for research. Modern deep learning methods have achieved good accuracy in violence detection and proved to be successful because of their applicability in intelligent surveillance systems. However, the models are computationally expensive and large in size because of their inefficient methods for feature extraction. This work presents a novel architecture for violence detection called Two-stream Multi-dimensional Convolutional Network (2s-MDCN), which uses RGB frames and optical flow to detect violence. Our proposed method extracts temporal and spatial information independently by 1D, 2D, and 3D convolutions. Despite combining multi-dimensional convolutional networks, our models are lightweight and efficient due to reduced channel capacity, yet they learn to extract meaningful spatial and temporal information. Additionally, combining RGB frames and optical flow yields 2.2% more accuracy than a single RGB stream. Regardless of having less complexity, our models obtained state-of-the-art accuracy of 89.7% on the largest violence detection benchmark dataset.