Visual Voice Activity Detection Research Articles

Learning spatiotemporal lip dynamics in 3D point cloud stream for visual voice activity detection

Visual voice activity detection (V-VAD) aims at identifying speech activities and measuring start/end moments in visual signal streams generated from the speaking-related biological modality. The visual cue is important counterpart of audio modality in noisy scenarios. Most of existing methods rely on 2D sequences, which suffer from limited robustness to variations of illumination and facial pose. This paper proposes a 3D V-VAD method based on lip dynamics modeling and learning spatiotemporal speaking lip representation in point cloud streams. We construct a lightweight temporal speaking dynamics model DualDyn to represent dual dynamics of global lip structure and local lip context in 4D spatiotemporal domain. The activity moment detection is formulated as frame-level state classification and refined by behavior continuity. Our method was verified on a public dataset S3DFM and achieves a state-of-the-art accuracy of 22.6ms and higher efficiency in measuring activity moments. Ablation studies demonstrate the robustness and applicability of our method.

Audiovisual speaker indexing for Web-TV automations

The current paper introduces a multimodal framework to provide Web-TV automations for live broadcasting and overall big streaming data management. The term indexing refers to the spatiotemporal localization of speakers participating in a discussion panel. Multiple modalities acting in parallel form the data-driven decision-making pipeline. The automated workflow includes the tasks of active speaker detection and localization, frame selection, and creation of a semantically annotated database. For improved performance and robustness, an information fusion model is proposed, which makes use of different audio and visual modalities. Audio-driven Voice Activity Detection follows the Enhanced Temporal Integration methodology applied on a standard audio feature set. For the localization of the dominant audio source, the argument that maximizes the General Cross-Correlation method is calculated. The visual modalities include face and mouth detection and Visual Voice Activity Detection. A Long Short Term Memory network is trained with mouth image sequences to determine voice activity. The values of the audio and visual Voice Activity Detection modules, as well as the General Cross-Correlation result, are used to train an Adaptive Neuro-Fuzzy model, which is responsible for the final decision. Experimental results prove the superiority of the information fusion approach compared to unimodal audio or visual models.

Using Spasmodic Closure Patterns to Simplify Visual Voice Activity Detection

While speaking, humans exhibit a number of recognizable patterns; most notably, the repetitive nature of mouth movement from closed to open. The following paper presents a novel method to computationally determine when video data contains a person speaking through the recognition and tally of lip facial closures within a given interval. A combination of Haar-Feature detection and eigenvectors are used to recognize when a target individual is present, but by detecting and quantifying spasmodic lip movements and comparing them to the ranges seen in true positives, we are able to predict when true speech occurs without the need for complex facial mappings. Although the results are within a reasonable accuracy range when compared to current methods, the comprehensibility and simple nature of the approach used can reduce the strenuousness of current techniques and, if paired with synchronous audio recognition methods, can streamline the future of voice activity detection as a whole.

Visual Voice Activity Detection in the Wild

The visual voice activity detection (V-VAD) problem in unconstrained environments is investigated in this paper. A novel method for V-VAD in the wild, exploiting local shape and motion information appearing at spatiotemporal locations of interest for facial video segment description and the bag of words model for facial video segment representation, is proposed. Facial video segment classification is subsequently performed using the state-of-the-art classification algorithms. Experimental results on one publicly available V-VAD dataset denote the effectiveness of the proposed method, since it achieves better generalization performance in unseen users, when compared to the recently proposed state-of-the-art methods. Additional results on a new unconstrained dataset provide evidence that the proposed method can be effective even in such cases in which any other existing method fails.

음성인식기 성능 향상을 위한 영상기반 음성구간 검출 및 적응적 문턱값 추정

In this paper, we propose an algorithm for achieving robust Visual Voice Activity Detection (VVAD) for enhanced speech recognition. In conventional VVAD algorithms, the motion of lip region is found by applying an optical flow or Chaos inspired measures for detecting visual speech frames. The optical flow-based VVAD is difficult to be adopted to driving scenarios due to its computational complexity. While invariant to illumination changes, Chaos theory based VVAD method is sensitive to motion translations caused by driver`s head movements. The proposed Local Variance Histogram (LVH) is robust to the pixel intensity changes from both illumination change and translation change. Hence, for improved performance in environmental changes, we adopt the novel threshold estimation using total variance change. In the experimental results, the proposed VVAD algorithm achieves robustness in various driving situations.

Voice activity detection based on facial movement

We present and evaluate a new Visual Voice Activity Detection method based on Spatiotemporal Gabor filters (STem-VVAD). Since Spatiotemporal Gabor filters are dynamic, they offer an attractive method to separate speech from non-speech frames in video, even though they have not been used for this purpose before. We evaluate our method on two datasets, which differ in the ratio of speech to non-speech frames (high versus low), as well as in the head orientation of the speakers (frontal versus profile). We compare models on different regions (applied to the mouth, the head or the entire video frame), and do so both for speaker-dependent, individual models and speaker-independent, generic models. In general, best performances are obtained for speaker-dependent STem-VVAD applied to the mouth region, and combining information from different speeds. In all these cases, the system outperforms two reference systems, relying on frame differencing and static Gabor filters respectively, showing that Spatiotemporal Gabor filters indeed are beneficial for visual voice detection.

Interference Reduction in Reverberant <newline/>Speech Separation With Visual <newline/>Voice Activity Detection

© 2014 IEEE.The visual modality, deemed to be complementary to the audio modality, has recently been exploited to improve the performance of blind source separation (BSS) of speech mixtures, especially in adverse environments where the performance of audio-domain methods deteriorates steadily. In this paper, we present an enhancement method to audio-domain BSS with the integration of voice activity information, obtained via a visual voice activity detection (VAD) algorithm. Mimicking aspects of human hearing, binaural speech mixtures are considered in our two-stage system. Firstly, in the off-line training stage, a speaker-independent voice activity detector is formed using the visual stimuli via the adaboosting algorithm. In the on-line separation stage, interaural phase difference (IPD) and interaural level difference (ILD) cues are statistically analyzed to assign probabilistically each time-frequency (TF) point of the audio mixtures to the source signals. Next, the detected voice activity cues (found via the visual VAD) are integrated to reduce the interference residual. Detection of the interference residual takes place gradually, with two layers of boundaries in the correlation and energy ratio map. We have tested our algorithm on speech mixtures generated using room impulse responses at different reverberation times and noise levels. Simulation results show performance improvement of the proposed method for target speech extraction in noisy and reverberant environments, in terms of signal-to-interference ratio (SIR) and perceptual evaluation of speech quality (PESQ).

Visual voice activity detection via chaos based lip motion measure robust under illumination changes

In this paper, a vision based voice activity detection (VVAD) algorithm is proposed using chaos theory. In conventional VVAD algorithm, the movement measure of lip region is found by applying an optical flow algorithm to detect the visual speech frame using a motion based energy feature set. However, since motion based feature is unstable under illumination changes, a new form of robust feature set is desirable. It is propositioned that contextual changes such as lip opening or closing motion during speech utterances under illumination variation can be observed as chaos-like and the resultant complex fractal trajectories in phase space can be observed. The fractality is measured in phase space from two sequential video input frames and subsequently any visual speech frames are robustly detected. Representative experiments are performed in image sequence containing a driver scene undergoing illumination fluctuations in moving vehicle environment. Experimental results indicate that a substantial improvement is obtained in terms of achieving significantly lower false alarm rate over the conventional method.

Audiovisual Voice Activity Detection Based on Microphone Arrays and Color Information

Audiovisual voice activity detection is a necessary stage in several problems, such as advanced teleconferencing, speech recognition, and human-computer interaction. Lip motion and audio analysis provide a large amount of information that can be integrated to produce more robust audiovisual voice activity detection (VAD) schemes, as we discuss in this paper. Lip motion is very useful for detecting the active speaker, and in this paper we introduce a new approach for lips and visual VAD. First, the algorithm performs skin segmentation to reduce the search area for lip extraction, and the most likely lip and non-lip regions are detected using a Bayesian approach within the delimited area. Lip motion is then detected using Hidden Markov Models (HMMs) that estimate the likely occurrence of active speech within a temporal window. Audio information is captured by an array of microphones, and the sound-based VAD is related to finding spatio-temporally coherent sound sources through another set of HMMs. To increase the robustness of the proposed system, a late fusion approach is employed to combine the result of each modality (audio and video). Our experimental results indicate that the proposed audiovisual approach presents better results when compared to existing VAD algorithms.

입술 영역의 움직임과 밝기 변화를 이용한 음성구간 검출 알고리즘 개발

음성구간을 검출하는 일반적인 방법은 음향신호로부터 특징값을 추출하여 판별식을 거치는 것이다. 그러나 잡음이 많은 환경에서 그 성능은 당연히 저하되며, 이 경우 영상신호를 이용하거나 영상과 음성을 동시에 사용함으로써 성능향상을 도모할 수 있다. 영상신호를 이용하여 음성구간을 검출하는 기존 방법들에서는 액티브 어피어런스 모델, 옵티컬 플로우, 밝기 변화 등 주로 하나의 특징값을 이용하고 있다. 그러나 음성구간의 참값은 음향신호에 의해 결정되므로 한 가지의 영상정보만으로는 음성구간을 검출하는데 한계를 보이고 있다. 본 논문에서는 입술 영역의 옵티컬 플로우와 밝기 변화 두 가지 영상정보로부터 특징값을 추출하고, 추출된 특징값들을 결합하여 음성구간을 검출하는 알고리즘을 제안하고자 한다. 또한, 음성구간 검출 알고리즘이 다른 시스템의 전처리로 활용되는 경우에 적은 계산량만으로 수행되는 것이 바람직하므로, 통계적 모델링에 의한 방법보다는 추출된 특징값으로부터 간단한 대수적 연산만으로 스코어를 산정하여 문턱값과 비교하는 방법을 제안하고자 한다. 입술 영역 검출을 위해서는 얼굴에서 가장 두드러진 특징점을 갖는 눈을 먼저 검출한 후, 얼굴의 구조와 밝기값을 이용하는 알고리즘을 제안하였다. 실험 결과 본 논문에서 제안하는 두 가지 특징값을 결합한 음성구간 검출 알고리즘이 하나의 특징값만을 이용했을 때보다 우수한 성능을 보임을 확인할 수 있다. Voice activity detection (VAD) is generally conducted by extracting features from the acoustic signal and a decision rule. The performance of such VAD algorithms driven by the input acoustic signal highly depends on the acoustic noise. When video signals are available as well, the performance of VAD can be enhanced by using the visual information which is not affected by the acoustic noise. Previous visual VAD algorithms usually use single visual feature to detect the lip activity, such as active appearance models, optical flow or intensity variation. Based on the analysis of the weakness of each feature, we propose to combine intensity change measure and the optical flow in the mouth region, which can compensate for each other's weakness. In order to minimize the computational complexity, we develop simple measures that avoid statistical estimation or modeling. Specifically, the optical flow is the averaged motion vector of some grid regions and the intensity variation is detected by simple thresholding. To extract the mouth region, we propose a simple algorithm which first detects two eyes and uses the profile of intensity to detect the center of mouth. Experiments show that the proposed combination of two simple measures show higher detection rates for the given false positive rate than the methods that use a single feature.

Robust visual speakingness detection using bi-level HMM

Visual voice activity detection (V-VAD) plays an important role in both HCI and HRI, affecting both the conversation strategy and sync between humans and robots/computers. The typical speakingness decision of V-VAD consists of post-processing for signal smoothing and classification using thresholding. Several parameters, ensuring a good trade-off between hit rate and false alarm, are usually heuristically defined. This makes the V-VAD approaches vulnerable to noisy observation and changes of environment conditions, resulting in poor performance and robustness to undesired frequent speaking state changes. To overcome those difficulties, this paper proposes a new probabilistic approach, naming bi-level HMM and analyzing lip activity energy for V-VAD in HRI. The designing idea is based on lip movement and speaking assumptions, embracing two essential procedures into a single model. A bi-level HMM is an HMM with two state variables in different levels, where state occurrence in a lower level conditionally depends on the state in an upper level. The approach works online with low-resolution image and in various lighting conditions, and has been successfully tested in 21 image sequences (22,927 frames). It achieved over 90% of probabilities of detection, in which it brought improvements of almost 20% compared to four other V-VAD approaches.

Visual voice activity detection with optical flow

Current voice activity detection methods generally utilise only acoustic information. Therefore they are susceptible to false classification because of the presence of other acoustic sources such as another speaker or non-stationary noise. To address this issue, the authors propose a new method of voice activity detection using solely visual information in the form of a speaker's mouth region. Such video information is not affected by the acoustic environment. Simulations show that a high percentage correct silence detection (CSD) can be obtained with a low percentage false silence detection (FSD). Comparisons with two other visual voice activity detectors show the proposed method to be consistently more accurate, and on average yields a 4% improvement in CSD. The usefulness of the method is confirmed by applying it to a previously published audio–visual convolutive blind source separation algorithm, to increase the intelligibility of a speaker.

A study of lip movements during spontaneous dialog and its application to voice activity detection

This paper presents a quantitative and comprehensive study of the lip movements of a given speaker in different speech/nonspeech contexts, with a particular focus on silences (i.e., when no sound is produced by the speaker). The aim is to characterize the relationship between "lip activity" and "speech activity" and then to use visual speech information as a voice activity detector (VAD). To this aim, an original audiovisual corpus was recorded with two speakers involved in a face-to-face spontaneous dialog, although being in separate rooms. Each speaker communicated with the other using a microphone, a camera, a screen, and headphones. This system was used to capture separate audio stimuli for each speaker and to synchronously monitor the speaker's lip movements. A comprehensive analysis was carried out on the lip shapes and lip movements in either silence or nonsilence (i.e., speech+nonspeech audible events). A single visual parameter, defined to characterize the lip movements, was shown to be efficient for the detection of silence sections. This results in a visual VAD that can be used in any kind of environment noise, including intricate and highly nonstationary noises, e.g., multiple and/or moving noise sources or competing speech signals.

Visual voice activity detection as a help for speech source separation from convolutive mixtures

Audio–visual speech source separation consists in mixing visual speech processing techniques (e.g., lip parameters tracking) with source separation methods to improve the extraction of a speech source of interest from a mixture of acoustic signals. In this paper, we present a new approach that combines visual information with separation methods based on the sparseness of speech: visual information is used as a voice activity detector (VAD) which is combined with a new geometric method of separation. The proposed audio–visual method is shown to be efficient to extract a real spontaneous speech utterance in the difficult case of convolutive mixtures even if the competing sources are highly non-stationary. Typical gains of 18–20 dB in signal to interference ratios are obtained for a wide range of (2 × 2) and (3 × 3) mixtures. Moreover, the overall process is computationally quite simpler than previously proposed audio–visual separation schemes.