Shigeki Miyabe

Temporal quantization of spatial information using directional clustering for multichannel audio coding

Binaural cue coding, which is a representing low bit-rate coding of multichannel audio, generates large distortion when the audio data have complex spatial image, such as symphony. Such distortion caused by the low frequency resolution of spatial information because BCC quantizes the parameters of localization. In this paper we propose a new coding framework by quantizing the spatial information temporally. The single-channel sum signal is panned to the multiple channels by selecting the prototypes of the spatial filter. Optimization of the prototypes with minimum coding error is given by a k-means-like clustering of the angles whose centroids are given by the first principal components of the covariances in the classes. The efficiency of the proposed coding with high quality is verified both in the objective and subjective evaluations.

Blind estimation of locations and time offsets for distributed recording devices

This paper presents a blind technique to estimate locations and recording time offsets of distributed recording devices from asynchronously recorded signals. In our method, locations of sound sources and recording devices, and the recording time offsets are estimated from observed time differences of arrivals (TDOAs) by decreasing the mean squared errors. The auxiliary-function-based updates guarantee the monotonic decrease of the objective function at each iteration. The TDOAs are estimated by the generalized cross correlation technique. The validity of our approach is shown by experiments in real environment, where locations of seven sound sources and eight microphones and eight time offsets were estimated from signals recorded by four stereo IC recorders in reverberant rooms.

Auxiliary-function-based independent component analysis for super-Gaussian sources

This paper presents new algorithms of independent component analysis (ICA) for super-Gaussian sources based on auxiliary function technique. The algorithms consist of two alternative updates: 1) update of demixing matrix and 2) update of weighted covariance matrix, which include no tuning parameters such as step size. The monotonic decrease of the objective function at each update is guaranteed. The experimental results show that the derived algorithms are robust to nonstationary data and outliers, and the convergence is faster than natural-gradient-based algorithm.

Automatic video annotation via Hierarchical Topic Trajectory Model considering cross-modal correlations

We propose a new statistical model, named Hierarchical Topic Trajectory Model (HTTM), for acquiring a dynamically changing topic model that represents the relationship between video frames and associated text labels. Model parameter estimation, annotation and retrieval can be executed within a unified framework with a few computation. It is also easy to add new modals such as audio signal and geotags. Preliminary experiments on video annotation task with manually annotated video dataset indicate that our proposed method can improve the annotation accuracy.

Audio object individual operation and its application to earphone leakage noise reduction

In this paper, we propose a new extension framework of multichannel audio coding based on temporal quantization of spatial information. In our previous study, multiple-audio-object signal can be encoded/decoded via prototypes of directional clustering for each audio object. This paper, first, pays attention to the fact that quantized information corresponds to the spatial image of each sound object, and is easily modified and operated by the end users. Next, as a novel application of the spatial image operation, we propose a new earphone leakage noise reduction method in which the mostly leaked sound object can be detected automatically based on the higher-order statistics, and eliminated. The effectiveness of the proposed CODEC is verified in the subjective evaluation, and leakage noise reduction method is assessed in the objective evaluation.

Analysis on speech characteristics for robust voice activity detection

This paper discusses about effective speech characterization for off-line voice activity detection (VAD), which is an important step prior to speech data mining. Five different natures of speech are examined; energy, spectral shape, periodicity, phonetic variation, and spectral fluctuation, the latter observed from a new point of view. Specific spectral fluctuation patterns of speech have been analyzed using multi-stage Harmonic/Percussive Sound Separation algorithm. We compared the performance of the features, and various combinations, to evaluate their robustness in multiple noise environments. The combined approach outperformed the baseline of CENSREC-1-C evaluation framework. The results suggest that the proposed feature extraction approach can improve state of the art VAD methods.

On the non-uniqueness problem and the semi-blind source separation

Semi-blind source separation (SBSS) is a special case of the well-known source separation problem when some partial knowledge of the source signals is available to the system. In particular, a batch-wise adaptation in the frequency domain based on the independent component analysis (ICA) can be effectively used to jointly perform source separation and multi-channel acoustic echo cancellation (MCAEC) without double-talk detection. However, the non-uniqueness problem due to the correlated far-end reference signals still affects the SBSS approach. In this paper, we analyze the structure of the SBSS de-mixing matrix and the behavior of a batch on-line adaptation algorithm under two most common far-end mixing conditions. We show that with a proper constraint on the de-mixing matrix, high echo reduction can be achieved just as the misalignment remains relatively low even for the worst-case scenario of single far-end talker and also without any pre-processing procedure to decorrelate the far-end signals.