Stop ‘them’ from spying on you: New AI can prevent rogue microphones TOU

Stop ‘them’ from spying on you: New AI can prevent rogue microphones

 TOU

Stop ‘them’ from spying on you: New AI can prevent rogue microphones

Credit: Unsplash / CC0 public domain

Have you ever noticed that very close online ads follow you on some of the things you talk about recently with your friends and family? Today microphones are embedded in almost everything from our phones, watches and televisions to voice assistants, and they will always listen to you. Computers constantly use neural networks and AI to process information about you. If you want to prevent this from happening, how can you do it?

In those days, you could turn up the volume and play music or turn on the water in the bathroom, as depicted on the hit TV show “The Americans”. But what if you don’t want to constantly scream at music? Columbia engineering researchers have developed a new system that creates whisper-quiet sounds that can prevent smart devices from spying on you in any room, in any situation. Also, it is easy to implement in hardware like computers and smartphones, which helps people in protecting the privacy of their voice.

“One of the key technical challenges in achieving this is to make everything work faster,” said Carl Wondrick, an assistant professor of computer science. “Our algorithm prevents the rogue microphone from hearing your words 80% of the time, which is fast and very accurate in our test bed. It works even when we know nothing about the rogue microphone. Hides from listening systems and makes conversation between people in the room easier. ”

Being in front of conversations

Although the group’s decision to dismantle automated speech recognition systems has been known in theory to be possible in AI for some time, achieving sufficient speed to apply them in practical applications remains a major obstacle. The problem is, now the sound of a person breaking his speech – at this particular moment – is not the sound of breaking his speech a second later. When people speak, their voices change constantly because they say different words and speak very fast. These changes make it impossible for a machine to keep up with a person’s fast speech speed.

“Our algorithm can proceed by predicting the characteristics of what a person will say next and giving them enough time to create the right gossip,” said Mia Sequier, the study’s lead author and Ph.D. Student at the Wondrick Lab. “So far our system works for most English vocabulary, and we plan to use the algorithm in multiple languages, as well as eventually making the whisper sound completely incomprehensible.”

Launch of ‘Predictive Attacks’

Researchers need to design a mechanism by which neural networks can be broken down in real time, which will be constantly developed during speech and apply to most vocabulary in a language. Previous work has successfully met one of these three requirements, but none have met all three. Chiquier’s new method uses what he calls “predictive attacks” —a signal that disrupts any word trained to translate automated speech recognition models. In addition, when attacking sounds are heard in the air, they should be loud enough to disrupt any rude “listening” microphone in the distance. The attacking sound should carry the same distance as the voice.

The researchers’ approach achieves real-time effectiveness by predicting a future attack of a signal or word based on two seconds of input speech. The panel is optimized for attack, so it has a level of sound similar to normal background noise, allowing those in a room to converse naturally and not being successfully monitored by the automatic speech recognition system. The team successfully demonstrated that their method works within real world rooms with natural ambient noise and complex visual geometry.

Protocol AI

“For many of us in the research community, ethical concerns about AI technology are an essential issue, but it seems to belong to a separate thought process. We & # 39; d really enjoyed creating a driving car, but forgot about design. Steering and brakes,” said a professor of computer and information science at the University of Pennsylvania. Jianpo Shi, a leading researcher in machine learning. “As a society, we need to ‘consciously’ think about the human and social impact of AI technology that we have developed from the early research design stage. Mia Chiquier and Carl Vondrick’s study raises the question of ‘how to use AI to protect ourselves unexpectedly. Their work makes many of us think in the following direction: Do not ask what protocol AI can do for us, but what can we do for protocol AI? If we believe in this direction, protocol AI research will be fun and creative. ”

Chiquier presents his paper at the International Conference on Learning Representatives on April 25, 2022.


                                    <div class="article-main__explore my-4 d-print-none">







                                        Machine learning enhances human speech recognition
                                    </div>                                      
                                    <hr class="mb-4"/>
                                                                                                                        <div class="article-main__more p-4">
                                                                                            <strong>More info:</strong>
                                            Mia Chiquier et al, Real-Time Neural Voice Camouflage (2022).  Available as a PDF at arXiv: 2112.07076 [cs.SD] <a href="https://arxiv.org/abs/2112.07076" target="_blank" rel="noopener">arxiv.org/abs/2112.07076</a>

                                                                                        </div>

                                                                                <div class="d-inline-block text-medium my-4">



                                            Presented by Columbia University School of Engineering and Applied Science
                                                                                                    <a class="icon_open" href="http://www.engineering.columbia.edu/" target="_blank" rel="nofollow noopener">
                                                    <svg>
                                                        <use href="https://techx.b-cdn.net/tmpl/v2/img/svg/sprite.svg#icon_open" x="0" y="0"/>
                                                    </svg>
                                                </a>

                                        </div>

                                    <!-- print only -->
                                    <div class="d-none d-print-block">
                                        <p>
                                             <strong>Quote</strong>: Stop 'them' from spying on you: new AI can prevent rogue microphones (2022, April 18) https://techxplore.com/news/2022-04-spying-ai-block-rogue-microphones from April 18, 2022 Retrieved from .html





                                        This document is subject to copyright.  No part may be reproduced without written permission, except for any reasonable manipulation for the purpose of personal study or research.  Content is provided for informational purposes only.

                                    </div>

                                </div>https://connect.facebook.net/en_US/sdk.js</p>

if you want to read this article from the original credit source of the article then you can read from here

.

Leave a Reply