In the future, it will be possible to integrate the algorithm into equipment for the hearing impaired and allow them to focus their hearing on the point they are looking at.
One of the ways that the planners of attacks or the people of organized crime of various kinds use to communicate with each other without fear of the police eavesdropping, is to carry out these conversations in rooms crowded with people, so that even if a microphone picks them up it will be very difficult to decipher things, mainly due to the huge background noise of the other conversations which are carried out around them at the same time.
At the science fair held recently at Bar-Ilan University, Dr. Sharon Ganot from the School of Engineering presented a system for extracting speakers from a night of conversations.
"The new algorithm uses a number of microphones scattered in space," explains Dr. Genot. "First we locate the desired speakers by scanning the space. The information reaching the array of microphones is filtered using a spatial filter we developed that succeeds in deleting the unwanted speakers and leaving the desired speakers almost without distortion. In doing so, we were able to, in fact, deal with a phenomenon known as the "cocktail party problem", where many people are talking but we want to hear the conversation of one couple out of all of them," says Ganut. According to him, the algorithm is flexible and after its initial activation can be mapped and listened to other speakers in the room space.
"In the next step, it will be possible to integrate the algorithm into hearing aids and thus help the hearing impaired when they are in a noisy environment," explains Dr. Genot. "For example, it will be possible to allow them to listen to the conversation taking place in the direction they are looking. Of course, when the police want to eavesdrop on suspects such as drug dealers who often meet in crowded places, it will be possible to listen to them, if it is possible for their people to plant several microphones," adds Dr. Ganot. According to him, Bar-Ilan is now working on perfecting the algorithm with the help of funded research. Among other things, the goal is to follow the speakers who move around the room and do not stand in one place. In this framework, a unique acoustic laboratory in Israel is being established in the School of Engineering, which will allow experiments to be carried out under real conditions.
This development is part of Shmulik Markowitz's master's thesis under the guidance of Dr. Sharon Ganot and Prof. Israel Cohen from the Technion.
11 תגובות
Basing only on time differences will not work at high frequencies because of the ambiguity phenomenon, so it will be necessary to use other parameters besides the ITD
I think the problem they approached in the article is the fact that there are more conversations from microphones and therefore a single solution from the ICA is not warranted.
The other way is to use an acoustic model of the environment to add information to ICA and calculate what the weight vector of all the microphones is, called: forwarded model
It's a bit unclear to me what's new here, the acoustic model? Methods are used as if neuroscience has been around for a long time: trying to "listen" to groups of neurons from the entire brain using "microphones of the type:" EEG or MEG
A. Is there not a violation here of the law that prohibits eavesdropping on a conversation that the quoted person is not a party to, except by order of the High Court?
B. I can think of two possible detection methods. One is locating the modulation of the particular speech that is being sought. This of course requires a prior sampling of the quoted person. The second is precise location in space of the quoted person, for example by a laser beam and then, the direction of the microphone system for focused listening from that point. This may damage the secrecy of the quotation.
jubilee:
It is definitely relevant (
http://en.wikipedia.org/wiki/Independent_Component_Analysis
) but it is likely that the solution that was implemented somehow also refers to the time differences (at least in the sense of taking steps to neutralize them) because otherwise there would be no need to adjust the system to the state of the sound source's movement.
Besides, they clearly talk about a spatial filter.
Independent Component Analysis (ICA)
The theory has been around for about 10 years
I actually liked the fact that the device was developed for scientific needs and to help the hard of hearing, however the article talks about the world of crime and intelligence.
Turning to the defense market multiplies the chance that the technology will be realized, the fact that funding has been received for further development and they will build one of the largest acoustic laboratories in the country.
Sometimes you also need to know how (and to whom) to market the product in order to leverage advanced technologies.
Ami:
It will not bother the peaceful citizen. It will only disturb those who make a sound 🙂
point:
By the way, it is most likely that DTOA is one of the components of the system, but it is also likely that it is not the Kaltz vision
"Among other things, the goal is to follow the speakers who move around the room and do not stand in one place"
This is a piece of information that could have been spared from the world of crime. Now they know how to dodge too easily.
Very interesting research. All the best and I hope that it will soon reach an applied state for both the police and the hearing impaired.
The scary situation is that now, with such an algorithm, the state will start spreading microphones everywhere and the element of intrusion into the privacy of the quiet and law-abiding citizen will rise another level. It is scary and unpleasant to know that you are potentially being listened to. I hope they knew how to use development for good always.
Greetings friends,
Ami Bachar
point:
you are missing the point :)
You can find the position using time differences only when you recognize the same sound in several microphones but if you are able to recognize "the same sound" then you have already solved the problem.
It doesn't seem that complicated. Using time differences that can be measured with great precision using a computer, the source of the sound can be located in space. I guess the first step is some kind of system calibration.