Search

Source separation involving mono-channel audio is a challenging problem, in particular for speech separation where source contributions overlap both in time and frequency. This task is of high interest for applications such as video conferencing. Recent progress in machine learning has shown that the combination of visual cues, coming from the video, can increase the source separation performance. Starting from a recently designed deep neural network, we assess its ability and robustness to separate the visible speakers’ speech from other interfering speeches or signals. We test it for different configuration of video recordings where the speaker’s face may not be fully visible. We also asses the performance of the network with respect to different sets of visual features from the speakers’ faces.

Estimation of the location of sound sources is usually done using microphone arrays. Such settings provide an environment where we know the difference between the received signals among different microphones in the terms of phase or attenuation, which enables localization of the sound sources. In our solution we exploit the properties of the room transfer function in order to localize a sound source inside a room with only one microphone. The shape of the room and the position of the microphone are assumed to be known. The design guidelines and limitations of the sensing matrix are given. Implementation is based on the sparsity in the terms of voxels in a room that are occupied by a source. What is especially interesting about our solution is that we provide localization of the sound sources not only in the horizontal plane, but in the terms of the 3D coordinates inside the room.

Digitalization of television signal has brought many changes not only to the way that data is presented, but also to amount of data that is being transferred. Transport stream provides user with big number of programs and very detailed information about every day's schedule. Aim of our solution is to present user the most appropriate TV shows from schedule and to help user to avoid energy and time consuming browsing through huge amount of data. Our solution should meet users viewing habits and tastes. In this paper experience during implementation of a recommendation engine dedicated to television data is presented. Focus is on requirements and environment (framework) for recommendation of television data. Future plans are also presented.

Age of digital television has brought many changes to television user's experience. On one side there is comfort because of great amount of data available in the stream and possibility of accessing the Internet over TV. On the other side great amount of data brings significant discomfort, if the data is not organized well and user has to browse through it for a long time. This paper describes a solution for recommending similar data originating from digital television stream or from the Internet. Aim of the presented solution is personalization of television content and related data from the Internet and also selection of most proper results. Main approaches are described and the solution is compared to some existing solutions.

Digital television has been a standard way of distributing television content for some time now. Even though digital television standards (e.g. DVB) usually provide some mechanisms for the distribution of metadata related to the broadcasted content (e.g. Electronic Program Guide, teletext, HbbTV, etc.), this information is mostly incomplete, missing and difficult to manage. In this paper, we describe a cloud based solution that copes with these problems by expanding the basic DTV metadata obtained from the broadcasters with content freely available on the internet and by employing a recommendation engine to filter and sort the content before delivering it to the users. For this purpose, we developed an acquisition system for acquiring base DTV metadata, a search engine for base metadata expansion and a simple recommendation engine for per-user content selection.

In order to address optimal acoustic control in rooms, an accurate characterization of the room shape and wall properties is required. There are only a few approaches that model the wall impedances. Most of them rely on finite difference time domain methods, which are limited to shoebox-shaped rooms and only valid at low frequencies (non-rectangular rooms and high frequencies lead to extremely high computational complexity). In order to overcome these limits, we propose the estimation of walls’ acoustic impedances based on the analysis of the room impulse responses. In room impulse responses, the early reflections’ amplitudes are proportional to the reflection coefficient of the corresponding wall element. The location of points on the walls (origins) for the first and second order echoes can be easily determined for a known room. The values of impedances at the origins of the first order echoes are determined directly from the room impulse response. The second order echoes' represent a product of the inf...

In last decades a vast number of textual analyzers have been developed. All those analyzers can be applied to textual data from any source, including Digital Video Broadcasting (DVB) stream. In this paper we propose a realization of a system whose primary purpose is DVB stream data acquisition. This solution can be used on a big variety of set-top boxes (STBs), considering their performances and capabilities and it presents an essential starting point for numerous types of systems, including television systems, personalized television systems etc.

This paper presents one solution for a extendable application which is used to collect EPG (Electronic Program Guide) data from various internet sources. The aim of the software is to complement or replace incomplete EPG information received from DTV (Digital television) stream with Internet data. The acquired data is processed and forwarded to remote server which further stores it to a database.

We study the detailed temporal evolution of echo density in impulse responses for applications in acoustic analysis and rendering on general environments. For this purpose, we propose a smooth sorted density measure that yields an intuitive trend of echo density growth with time. This is fitted with a general power-law model motivated from theoretical considerations. We validate the framework against theory on simple room geometries and present experiments on measured and numerically simulated impulse responses in complex scenes. Our results show that the growth power of echo density is a promising statistical parameter that shows noticeable, consistent differences between indoor and outdoor responses, meriting further study.

We investigate the design of a convolutional layer where kernels are parameterized functions. This layer aims at being the input layer of convolutional neural networks for audio applications or applications involving time-series. The kernels are defined as one-dimensional functions having a band-pass filter shape, with a limited number of trainable parameters. Building on the literature on this topic, we confirm that networks having such an input layer can achieve state-of-the-art accuracy on several audio classification tasks. We explore the effect of different parameters on the network accuracy and learning ability. This approach reduces the number of weights to be trained and enables larger kernel sizes, an advantage for audio applications. Furthermore, the learned filters bring additional interpretability and a better understanding of the audio properties exploited by the network.