### Citations

949 |
Tracking and Data Association
- Bar-Shalom, Fortmann
- 1988
(Show Context)
Citation Context ...mate the parameters for association probabilities and state estimates. Another method that provides a solution for the data association problem is joint probabilistic data association filter (JPDAF)) =-=[57]-=-, which is an 1 The intractability depends (among others) on the assumptions about the state-space. However “real life” tracking problems are typically intractable due to the non-linear- and non-Gauss... |

651 |
The generalized correlation method for estimation of time delay
- Knapp, Carter
- 1976
(Show Context)
Citation Context ...s: conj(Sr) = real(Sr) − j imag(Sr), (2.11) where G(ω) is a weighting term (responsible for the pre-filtering). The phase transform (PHAT) weighting function was introPHAT weighting function duced in =-=[36]-=- and is defined as: G(ω) = |Sl(ω)S ∗ r (ω)| −1 . (2.12) 9This weighting function places equal importance on each frequency, by dividing the spectrum by its magnitude. By applying this whitening filte... |

486 |
Condensation:conditional density propagation for visual tracking
- Isard, Blake
- 1998
(Show Context)
Citation Context ...5 . 5 Note that in our experiments we requested the speakers to speak in turns. 88Appendix A Langevin Model There are several models that describe the (typical) dynamics of a person moving in a room =-=[30]-=-. The Langevin model can be used to represent the time-varying locations of a speaker. The Langevin equations are commonly used in the research field physics and chemistry to describe dynamic processe... |

251 | On Sequential Simulation-Based Methods for Bayesian Filtering,”
- Doucet
- 1998
(Show Context)
Citation Context ...describe the motion of the speaker. 2.2.1 Inference Methods Estimating the unknown process x based on sensory data y presented in the Bayesian framework implies solving the Bayesian filtering problem =-=[19]-=- . The posterior distribution p(x0, x1, . . . , xt|y0, y1, . . . , yt) includes all relevant information on {x0, x1, . . . , xt} at time t in the Bayesian framework. Typically, in signal processing ap... |

174 | People tracking with mobile robots using sample-based joint probabilistic data association filters
- Schulz, Burgard, et al.
(Show Context)
Citation Context ...g sections. First the background of the JPDAF is discussed, followed by a description of the general framework of the JPDAF. Next we discuss a generic sample-based version of the JPDAF as proposed by =-=[55]-=-. Finally, we present our model and (sample-based JPDAF) filtering algorithm for combining azimuth features with voice features. 4.1 Background The PDAF could be described as an extension of the Kalma... |

166 | Learning dynamic bayesian networks.
- Ghahramani
- 1998
(Show Context)
Citation Context ...describes a probabilistic dependency of measurements on the state p(λn|si n). Under such a framework will compute posterior state distribution p(si n|λ1:n) given measurements using Bayesian filtering =-=[27]-=-. On the basis of this distribution we associate segments with speakers. The state of ith person during the nth segment is described by si n = [αi n, T i ], where αi n = [yi n, ˙y i n, xi n, ˙x i n] i... |

155 | Tracking multiple moving targets with a mobile robot using particle filters and statistical data association.
- Schulz, Burgard, et al.
- 2001
(Show Context)
Citation Context ...) The probability of an individual joint association event conditioned on the measurement at time t can be derived with the use of Bayes‘ rule and the assumption that the estimation problem is Markov =-=[56]-=-. Thus we can compute p(Θ|Z t ) as follows: p(Θ|Z t ) = p(Θ|Z(t), Z t−1 ) (4.7) = = Markov! = Bayes! = ∫ ∫ ∫ ∫ p(Θ, X t |Z(t), Z t−1 ) dX t p(Θ|Z(t), Z t−1 , X t ) p(X t |Z(t), Z t−1 ) dX t p(Θ|Z(t), ... |

152 | An MCMC-based particle filter for tracking multiple interacting targets.
- Khan, Balch, et al.
- 2004
(Show Context)
Citation Context ...ated, given these assignment probabilities. Thus, a probabilistic exclusion principle 6 is accomplished by making the particle filters dependent through the evaluation of the assignment probabilities =-=[33]-=-. 3 With a naive approach, one could just choose a uniform distribution 4 Following from Gaussian motion distribution and a mixture-of-Gaussian sensor distribution, the filtered distribution takes the... |

130 | Speech recognition with dynamic Bayesian networks,
- Zweig
- 1998
(Show Context)
Citation Context ...pproximates the sophisticated human vowel detection and identification process. A more sophisticated method for speech recognition can be accomplished with the use of dynamic Bayesian networks (DBNs) =-=[67]-=-. DBNs can be used to represent complex stochastic processes. In the context of speech modelling, the DBNs provide us a convenient method that maintain an explicit representation of the lips, tongue, ... |

128 | The quantal nature of speech: Evidence from articulatory-acoustic data”, in - Stevens - 1972 |

127 | Experiences with a mobile robotic guide for the elderly.
- Montemerlo, Pineau, et al.
- 2002
(Show Context)
Citation Context ...s issue becomes more important, as current robots are moving out of the factory floor into environments inhabited by human. Examples are museum or exhibition robots [60], [1], care-for-elderly robots =-=[42]-=-, office [2] and entertainment robots [11]. In their role as guide, companion or servant these systems have to interact with the humans they encounter. Therefore, in robot perception and human-robot s... |

118 | Sequential Monte Carlo Methods for Multiple Target Tracking and Data Fusion
- Hue, Cadre, et al.
- 2002
(Show Context)
Citation Context ...ency extraction for each windowed data. 63Chapter 4 Multiple Target Tracking Tracking multiple moving targets in general requires estimating the joint probabilistic distribution of the target states =-=[29]-=-, [32]. As stated in chapter two, in practice computing the filtering distribution of the state of a single target is typically intractable 1 . Obviously, computing the filtering distribution of the s... |

100 | A robust method for speech signal time-delay estimation in reverberant rooms
- Brandstein, Silverman
- 1997
(Show Context)
Citation Context ...t signal since there usually is no a priori knowledge about the statistics of the involved signals. The PHAT localization algorithm is robust and reliable in realistic reverberant acoustic conditions =-=[10]-=-, however the algorithm does not take the signal to noise ratio (SNR) conditions into account. If the interference is dominant over the desired signal, then the phase information obtained from the PHA... |

84 | A practical methodology for speech source localization with microphone arrays.
- Brandstein, Silverman
- 1997
(Show Context)
Citation Context ..., one can increase the sensing resolution by using multiple microphone pairs [47]. With the use of an array of eight microphones one is able recover the three-dimensional location of the sound source =-=[8]-=-. From each microphone pair that is part of the microphone array we can recover a vector that indicates the sound source location. Thus from the centre of each microphone pair a vector can be formulat... |

81 | Nonlinear filtering for speaker tracking in noisy and reverberant environments,
- Vermaak, Blake
- 2000
(Show Context)
Citation Context ...ue sound source location. Furthermore, as stated earlier, the transformation from the TDOA estimates to source locations are nonlinear. As a consequence the state-space is non-Gaussian and non-linear =-=[62]-=-. As stated before, with these properties of the state-space, typically no-closed form solutions exist for computing the filtering distributions. Computing the filtering distribution in non-linear and... |

78 | Acoustic event localization using a crosspower-spectrum phase based technique. In:
- Omologo, Svaizer
- 1994
(Show Context)
Citation Context ...nique computes the TDOA as the inverse Fourier Transform of the received signal crosspower spectrum scaled by a weighting function. This method pre-filters the signal before computing the correlation =-=[48]-=-, [49]. Definition GCC Let s(t) denote the data (in the time domain) received at time t and let F{.} denote the Fourier transform and let S(ω) = F{s(t)} represent the data received at time t in the fr... |

69 | Use of the crosspower-spectrum phase in acoustic event location,”
- Omologo, Svaizer
- 1997
(Show Context)
Citation Context ...computes the TDOA as the inverse Fourier Transform of the received signal crosspower spectrum scaled by a weighting function. This method pre-filters the signal before computing the correlation [48], =-=[49]-=-. Definition GCC Let s(t) denote the data (in the time domain) received at time t and let F{.} denote the Fourier transform and let S(ω) = F{s(t)} represent the data received at time t in the frequenc... |

65 |
A theoretical framework for sequential importance sampling with resampling.
- Liu, Chen, et al.
- 2001
(Show Context)
Citation Context ...distribution before resampling. A Resampling Method A method for resampling is discussed here in more detail. First a measure is needed to estimate the number of near-zero-weight particles. Lui et al =-=[38]-=- refer to coefficient of variation cv2 t and is defined as: cv 2 t = var(wt,k) E 2 (wt,k) = 1 M M∑ (M wk − 1) 2 . (2.40) With equation (2.40) the effective sampling size ESSt can be computed and is de... |

64 | Robust sound source localization using microphone array on a mobile robot
- Valin, Michaud, et al.
(Show Context)
Citation Context ...sing mechanism takes various physical effects into account like the acoustic shadowing created by the head and the reflections of the sound by the two ridges running along the edges of the outer ears =-=[61]-=-. This ability enables humans to locate sound sources in three dimensional space. Several methods have been developed in order to mirror the human sound source location capability. Typically, one reli... |

39 | Multi-modal anchoring for human-robot interaction. Robotics and Autonomous Systems, - Fritsch, Kleinehagenbrock, et al. - 2003 |

38 |
Research on individuality features in speech waves and automatic speaker recognition techniques
- Furui
- 1986
(Show Context)
Citation Context ...LPC coefficients. More elaborate approaches use LPC-cepstral coefficients together with their regressions coefficients [40]. A comprehensive survey of speaker identification techniques is provided in =-=[23]-=-. The main underlying assumption of the LPC method is that speech is produced by a buzzer at the end of the tube. The glottis is responsible for producing this buzz, which is characterized by its pitc... |

36 | Real-time speaker tracking using particle filter sensor fusion,”
- Chen, Rui
- 2004
(Show Context)
Citation Context ... the audio signals to steer the camera toward the current speaker. Alternatively, in limited closed areas, the audio feature could help keep track of a person who disappears from robots field of view =-=[13]-=-. The tracking problem can be roughly divided into two sub-problems. Estimating the azimuth angle and estimating the formant frequencies. Both problems are approached in a straight forward manner. In ... |

34 | Pitch Extraction and Fundamental Frequency: History and Current Techniques,"
- Gerhard
- 2003
(Show Context)
Citation Context ...er partials. If the waveform is periodic, such as a vowel utterance, these partials are harmonically related 2 . Typically the frequency of the lowest partial of the waveform is referred to as the f0 =-=[24]-=-. There are different theories on how the human auditory system perceives pitch. In general the perception of pitch becomes more distinct whenever the relation between the partials becomes more harmon... |

30 |
A Text-Independent Speaker Recognition Method Robust Against Utterance Variations
- Matsui, Furui
- 1991
(Show Context)
Citation Context ...f the spectral envelope. In the simplest case these features are taken to be the LPC coefficients. More elaborate approaches use LPC-cepstral coefficients together with their regressions coefficients =-=[40]-=-. A comprehensive survey of speaker identification techniques is provided in [23]. The main underlying assumption of the LPC method is that speech is produced by a buzzer at the end of the tube. The g... |

29 | Jijo-2: An office robot that communicates and learns
- Asoh, Motomura, et al.
(Show Context)
Citation Context ...es more important, as current robots are moving out of the factory floor into environments inhabited by human. Examples are museum or exhibition robots [60], [1], care-for-elderly robots [42], office =-=[2]-=- and entertainment robots [11]. In their role as guide, companion or servant these systems have to interact with the humans they encounter. Therefore, in robot perception and human-robot social intera... |

24 | Active speech source localization by a dual coarse-to-fine search
- Duraiswami, Zotkin, et al.
- 2001
(Show Context)
Citation Context ...ins a challenging task to accurately track a moving sound source in the presence of a strong multipath and noise sources. Furthermore the inverse map from TDOA to location is non-linear and ill-posed =-=[20]-=-, [66] (see appendix C for details). Therefore 4 Pitch is briefly discussed in section D.3 5 This property is briefly discussed in section D.1 11it is difficult to accurately track the azimuth with a... |

24 |
On decomposing speech into modulated components
- Rao, Kumaresan
(Show Context)
Citation Context ...ope of the observed speech segment is a method that involves the adaptive band-pass filterbank. The most part of this section describes the work of [43]. The author of [43] contributed to the work of =-=[51]-=- by adding additional adaptive components to the adaptive filterbank. We have modified the adaptive filterbank in order to construct the DFT 7 component in our formant frequency extraction algorithm. ... |

21 |
A Navigation Framework for Multiple Mobile Robots and its Application at the Expo.02 Exhibition
- Arras, Philippsen, et al.
- 2003
(Show Context)
Citation Context ... human-robot interaction. This issue becomes more important, as current robots are moving out of the factory floor into environments inhabited by human. Examples are museum or exhibition robots [60], =-=[1]-=-, care-for-elderly robots [42], office [2] and entertainment robots [11]. In their role as guide, companion or servant these systems have to interact with the humans they encounter. Therefore, in robo... |

20 | Robust global localization using clustered particle filtering,”
- Milstein, Sánchez, et al.
- 2002
(Show Context)
Citation Context ...ted in the belief that the speaker is at some location θ. In other words we want to obtain the posterior distribution of the state space (represented in azimuth angles) conditioned on the sensor data =-=[41]-=-. We can denote this belief by: Bel(θt) = p(θt|θ z t , . . . , θ z 0). (2.18) Where θt represents the state of the speaker at time t and θ z t denotes the location obtained by transforming the measure... |

19 |
Accelerated speech source localization via hierarchical search of steer response power,”
- Zotkin, Duraiswami
- 2004
(Show Context)
Citation Context ...challenging task to accurately track a moving sound source in the presence of a strong multipath and noise sources. Furthermore the inverse map from TDOA to location is non-linear and ill-posed [20], =-=[66]-=- (see appendix C for details). Therefore 4 Pitch is briefly discussed in section D.3 5 This property is briefly discussed in section D.1 11it is difficult to accurately track the azimuth with a deter... |

17 | Real-time sound source localization and separation for robot audition
- Nakadai, Okuno, et al.
(Show Context)
Citation Context ... amount of corrupted input data for the formant frequency extraction algorithm can be reduced. Appropriate switching can be achieved by analysing the signals on interaural intensity differences (IID) =-=[45]-=-. As mentioned in the first chapter the human auditory system finds spatial information (amongst other inferences) by acoustic shadowing. Note that if the sound source is to the left side of the human... |

16 | Robust Formant Tracking for Continuous Speech with Speaker Variability.’ MS Thesis. McMaster Unv
- Mustafa
- 2003
(Show Context)
Citation Context ...mate the formant frequencies from the spectral envelope of the observed speech segment is a method that involves the adaptive band-pass filterbank. The most part of this section describes the work of =-=[43]-=-. The author of [43] contributed to the work of [51] by adding additional adaptive components to the adaptive filterbank. We have modified the adaptive filterbank in order to construct the DFT 7 compo... |

16 | New direct approaches to robust sound source localization. In:
- Rui, Florencio
- 2003
(Show Context)
Citation Context ...ereas applying cross-correlation in the frequency domain implies multiplication: Rslsr = sl ⊙sr = F −1 (SlS∗ r ). Thus applying cross-correlation in the frequency domain yields a complexity reduction =-=[54]-=-. If the windowed segment contains N samples the complexity of computing the cross-correlation using equation 2.7 is O(N 2 ), whereas computing the cross-correlation using equation 2.10 is O(Nlog2N). ... |

15 | Joint probabilistic techniques for tracking objects using multiple vision clues
- Rasmussen, Hager
- 1998
(Show Context)
Citation Context ..., we present our model and (sample-based JPDAF) filtering algorithm for combining azimuth features with voice features. 4.1 Background The PDAF could be described as an extension of the Kalman filter =-=[52]-=-. As described in section 2.2.3, the Kalman filter provides a method how to update the state of a single target given a measurement, that is: the measurement contains one observed feature at time step... |

14 | Acoustic source direction by hemisphere sampling,
- Birchfield, Gillmor
- 2001
(Show Context)
Citation Context ...sually corresponds to a change in the TDOA between the two signals. The TDOA can be transformed to an arriving angle, which corresponds to relative position of the sound source to the microphone pair =-=[6]-=-. Our localization function employs the above described concepts for sound source localization. 2.1.1 Azimuth Angle The azimuth angle θ specifies the direction of the sound wave when it strikes the mi... |

14 | A User-Interface Robot for Ambient Intelligent Environments,” in ASER 03 1st international workshop on advances in service robotics
- Breemen
(Show Context)
Citation Context ... robots are moving out of the factory floor into environments inhabited by human. Examples are museum or exhibition robots [60], [1], care-for-elderly robots [42], office [2] and entertainment robots =-=[11]-=-. In their role as guide, companion or servant these systems have to interact with the humans they encounter. Therefore, in robot perception and human-robot social interaction, it is becoming more imp... |

12 |
Sur les problmes aux drives partielles et leur signification physique
- Hadamard
- 1902
(Show Context)
Citation Context ...rue sound source peak with this implicit weighting. 91Appendix C Non-linear And Ill-posed problems In a famous paper by Jacques Hadamard published in 1902 discusses the notion of a wellposed problem =-=[28]-=-. The same author argued in an earlier paper published in 1901 that well-posed problems are physically important problems are both solvable and uniquely solvable. In the same paper he gave examples of... |

10 |
Acoustic source localization in three-dimensional space using cross-power spectrum phase. In:
- Svaizer, Matassoni, et al.
- 1997
(Show Context)
Citation Context ...tracking algorithm by increasing the sensing resolution with omni-directional microphones. Further progress can be accomplished by using a microphone array of eight or more microphones as proposed by =-=[46]-=-. In order to improve the azimuth estimates we have contributed the adaptive sigma algorithm. By relating the amount of coherence energy from the potential TDOA candidates obtained from the GCC functi... |

9 | Sound source localization in reverberant environments using an outlier elimination algorithm
- Jan, Flanagan
- 1996
(Show Context)
Citation Context ...ly more than one intersection point is simultaneously present, due to noise and calibration errors. To tackle this problem in this over-determined system, an outlier detector algorithm is proposed by =-=[31]-=- The incorrect vectors are detected and discarded, leaving only the vectors that are regarded as the most reliable ones giving one intersection point. Another approach to locate sound sources proposed... |

7 | Discriminative binaural sound localization
- Ben-Reuven, Singer
- 2002
(Show Context)
Citation Context ...ignal, then the phase information obtained from the PHAT function will be unreliable. Methods for estimating the noise spectra and incorporating them in the weighting functions are discussed in [10], =-=[5]-=-. Aiming to locate the speaker, one should aim for detecting the voice in the spectrum. Therefore the weighting function should emphasize the regions that are likely to contain voice components (that ... |

7 |
Fast mutual exclusion
- Maskell, Briers, et al.
- 2004
(Show Context)
Citation Context ...d thereby preventing two trackers locking onto the same target. In the particle filtering literature the concept of mutual exclusion is referred to as treating the joint event as a nuisance parameter =-=[39]-=-. The joint event is integrated out, with the result that the effects of all the joint events on the targets distribution are included. This concept is also used in the Kalman filtering framework unde... |

6 | Confidence scoring of time difference of arrival estimation for speaker localization with microphone arrays
- Bechler, Kroschel
- 2002
(Show Context)
Citation Context ...und and Fs denotes the sampling frequency. However, allowing time shifts outside the interval [−τmax , τmax] for possible TDOA candidates could be useful as a confidence measure for the found maximum =-=[4]-=-. Basically, if the (global) maximum of the cross-correlation function is found inside the interval [−τmax , τmax] while the cross-correlation is computed over the interval [−(τmax + n) , τmax + n] wh... |

6 | Computational auditory scene analysis exploiting speech-recognition knowledge, Proceedings of the IEEE Workshop on Applications of Signal Processing to Audio and Acoustics, Mohonk. Ellis, D.P.W. (in press), Modeling the auditory organization of speech - a
- Ellis
- 1997
(Show Context)
Citation Context ...o resemble the human auditory system with physiologically oriented models. Such models should approximate the human ability to interpret sound mixtures as the combination of distinct acoustic sources =-=[21]-=-. Even from this brief overview, it is obvious that the research 1 is inspired by the human sound source location capability, together with human sophisticating mechanism for voice recognition allowin... |

6 | Talker Localization and Speech Enhancement in a Noisy Environment using a Microphone Array based Acquisition System. EUROSPEECH
- Omologo, Svaizer
- 1993
(Show Context)
Citation Context ...are coming from the front or the back. In order to compensate for the high level of complexity of the human auditory system, one can increase the sensing resolution by using multiple microphone pairs =-=[47]-=-. With the use of an array of eight microphones one is able recover the three-dimensional location of the sound source [8]. From each microphone pair that is part of the microphone array we can recove... |

6 |
MINERVA: A TourGuide Robot that Learns. Kunstliche Intelligenz
- Thrun, Bennewitz, et al.
- 1999
(Show Context)
Citation Context ...atural human-robot interaction. This issue becomes more important, as current robots are moving out of the factory floor into environments inhabited by human. Examples are museum or exhibition robots =-=[60]-=-, [1], care-for-elderly robots [42], office [2] and entertainment robots [11]. In their role as guide, companion or servant these systems have to interact with the humans they encounter. Therefore, in... |

5 |
Binaural modeling and auditory scene analysis
- Bodden
- 1995
(Show Context)
Citation Context ... An alternative to the signal-processing-motivated algorithms for acoustic source localization is provided by the work of Auditory Scene Analysis (ASA) or Computational Auditory Scene Analysis (CASA) =-=[7]-=-. Inspired by the human ability to analyse the auditory environment, the localization of acoustic sources and the perception of speech simultaneously with just two receivers, CASA research aims to res... |

4 |
Physiological measures of the precedence effect and spatial release from masking in the cat inferior colliculus
- Litovsky, Lane, et al.
- 2001
(Show Context)
Citation Context ...mation on the precedence effect (PE). The PE is referred to as the observation that two sounds occurring in rapid succession are perceived as a single auditory object localized near the leading sound =-=[37]-=-. A substantial amount of our azimuth tracking results suffered from noise artefacts caused by reverberation. We believe that we could suppress the influence of these noise artefacts further by approp... |

3 |
Speech-based localization of multiple persons for an interface robot
- Klaassen, Zajdel, et al.
- 2005
(Show Context)
Citation Context ...e A = I. Furthermore, we discard matrix B from equation 2.28. In the rest of this section, we present the details of computing these features. We note that the presented model is based on the work of =-=[35]-=-. 4.5.1 Overview Our primary goal is association of the measured features λn with one of the speakers. For this purpose, we describe the ith speaker with a state variable si n, which summarizes the pe... |

3 |
Particle filtering algorithms for acoustic source localization
- Ward, Lehmann, et al.
- 2003
(Show Context)
Citation Context ...it is difficult to accurately track the azimuth with a deterministic approach. Typically a statespace approach overcomes the drawback of using only current sensor data to locate the true sound source =-=[64]-=-. The key to this approach is that the true sound source follows a dynamical motion model between consecutive measurements, whereas there is no temporal consistency in the spurious peaks. The involved... |

2 |
A Unified Joint Probabilistic Data Association Filter with Multiple Models
- Davey, Colegrove
- 2001
(Show Context)
Citation Context ...ns between the measured features {0, . . . , mt} and all possible sources {0, . . . , T } at time t. The event space can be partitioned such that every measured feature has a unique identified source =-=[16]-=-. The source can be either clutter or one of the targets. Since each target produces at most one measurement, no more than one measured feature can have the same source within such a partition. The jo... |

2 |
nd E. Milios Probabilistic cooperative localization and mapping
- Rekleitis, Dudek
- 1993
(Show Context)
Citation Context ...nction f(θ|T DOA) are used in order to compute the sensor model equations. Resampling A disadvantage of approximating the filtering distribution with samples is the depletion of the sample population =-=[53]-=-. The particles that explore possible movements of the source which have drifted far from the actual movement or the measured movement will have near-zero weights. These particles do not substantially... |

1 |
Two-point Correlation Measurements of Density
- Basse, Zoletnik, et al.
- 1999
(Show Context)
Citation Context ...ated on the correct TDOA caused by high coherence between the two signals at the corresponding lag. A linear phase shift in the crosspower spectrum corresponds to a time shift in the crosscorrelation =-=[3]-=-. Aiming to find the relative time shift τlr (see also equation (2.6)) such that the two compared signals have the maximum coherence 3 , we can write the crosspower spectrum of the two measured signal... |

1 |
Brandstein A pitch-based approach to time-delay estimation of reverberant speech
- S
- 1997
(Show Context)
Citation Context ...weighted cross-correlation function is defined as: Rl,r(τ) = ∫ ∞ −∞ w 2 e(ω)Sl(ω)S ∗ r (ω) |Sl(ω)S ∗ r (ω)| exp (jωτ) dω. (2.17) Similar to this approach is a pitch-based TDOA estimator introduced by =-=[9]-=- where knowledge about the pitch characteristics of speech are exploited 4 . The voice has a narrow bandwidth and is usually concentrated in lower regions of the frequency spectrum. Furthermore, the v... |

1 |
Soon Ong Learning and Regularization from Interpolation to Approximation Lecture Notes, WWW: asi.insa-rouen.fr/∼scanu
- Canu, C
(Show Context)
Citation Context ...e problem P is well-posed if its solution: - exists - is unique - is stable (||f − ft|| ≤ C||y − yt||) - the solution depends continuously on the data Definition: Ill-posed problem above requirements =-=[12]-=-. Problem P is ill-posed if its solution violates one of the Inverse Map From TDOA Estimate To Azimuth Angle With Hadamards notion of ill-posedness, many inverse problems exhibit such behaviour, as th... |

1 |
Organisation and Design of Autonomous Systems Reader Faculty
- Dam, Dev, et al.
- 2000
(Show Context)
Citation Context ... 2.2.3 Kalman Filter The Kalman filter is a set of equations that provides an estimate of the state of a dynamic system. The filter provides a recursive solution to the linear filtering problem [65], =-=[15]-=-. We cannot compute the distance to the sound source with the use of only two microphones. Therefore we cannot compute the exact location in the Cartesian coordinate system. However, we can compute th... |

1 | Silence as a Cue to Rhythm in the Analysis of Speech and Song
- Gerhard
(Show Context)
Citation Context ...xtraction of additional features. An obvious choice would be extracting consonant based information (formant frequency estimation), a less obvious choice is extracting rhythm patterns from the speech =-=[25]-=-. These rhythm patterns are based on silence cues and could complement the formant frequencies patterns estimated from vowel and consonant utterances for speaker identification. As stressed in earlier... |

1 |
de Grooth A simple model for Brownian motion leading to the Langevin equation Am
- G
- 1999
(Show Context)
Citation Context ...represent the time-varying locations of a speaker. The Langevin equations are commonly used in the research field physics and chemistry to describe dynamic processes with stochastic excitation forces =-=[26]-=- [63]. The Langevin model is reasonably simple, but has been shown to work well in practice [64]. In the context of the moving speaker, the velocity is prescribed as a random process, with properties ... |

1 | Ohnishi Self-organisation of a Sound Source Localization Robot by Perceptual Cycle - Nakashima, Mukai, et al. - 2002 |

1 |
Piccon and J.P. Chabard Two Attempts of Turbulence Modelling in Smoothed Particle Hydronamics Proceeding of Flow Modeling and Turbulence Measurement
- Violeau, S
- 2001
(Show Context)
Citation Context ...sent the time-varying locations of a speaker. The Langevin equations are commonly used in the research field physics and chemistry to describe dynamic processes with stochastic excitation forces [26] =-=[63]-=-. The Langevin model is reasonably simple, but has been shown to work well in practice [64]. In the context of the moving speaker, the velocity is prescribed as a random process, with properties fulfi... |

1 |
An Introduction to the Kalman Filter Annual Conference on Computer Graphics and Interactive Techniques
- Welch, Bishop
(Show Context)
Citation Context ...ution. 2.2.3 Kalman Filter The Kalman filter is a set of equations that provides an estimate of the state of a dynamic system. The filter provides a recursive solution to the linear filtering problem =-=[65]-=-, [15]. We cannot compute the distance to the sound source with the use of only two microphones. Therefore we cannot compute the exact location in the Cartesian coordinate system. However, we can comp... |