Slashdot Mirror
Alexa Scientists Claim Audio Watermarking Technique Nearing 100% Accuracy (venturebeat.com)
georgecarlyle76 brought our attention to Amazon's claim of an algorithm that "solves the 'second-screen problem' in real-time."
"Ever hear (no pun intended) of audio watermarking?" asks VentureBeat. It's the process of adding distinctive sound patterns identifiable to PCs, and it's a major way web video hosts, set-top boxes, and media players spot copyrighted tracks. But watermarking schemes aren't particularly reliable in noisy environments, like when the audio in question is broadcasted over a loudspeaker. The resulting noise and interference -- referred to in academic literature as the "second-screen" problem -- severely distorts watermarks, and introduces delays that detectors often struggle to reconcile. Researchers at Amazon, though, believe they've pioneered a novel workaround, which they describe in a paper newly published on the preprint server Arxiv ("Audio Watermarking over the Air with Modulated Self-Correlation") and an accompanying blog post. The team claims their method -- which they'll detail at the International Conference on Acoustics, Speech, and Signal Processing in May -- can detect watermarks added to about two seconds of audio with "almost perfect accuracy," even when the distance between the speaker and detector is greater than 20 feet...
So how's it work? As Tai explains, the model employs a "spread-spectrum" technique in which watermark energy is spread across time and frequency, rendering it inaudible to human ears while robustifying it against postprocessing (like compression). And it generates watermarks from noise blocks of a fixed duration, each of which introduces its own distinct pattern to selected frequency components in the host audio signal. Conventional detectors would compare the resulting sequence of noise blocks -- the decoding key -- with a reference copy. But Tai and colleagues take a different approach: Their algorithm embeds the noise pattern in the audio signal multiple times and compares it to itself. Because said signal passes through the same acoustic environment, Tai explains, instances of the pattern are distorted in similar ways, enabling them to be compared directly. "The detector takes advantage of the distortion due to the acoustic channel, rather than combatting it," he added.
"Audio content that Alexa plays -- music, audiobooks, podcasts, radio broadcasts, movies -- could be watermarked on the fly," explains Amazon's blog post. It argues that this could be useful "so that Alexa-enabled devices can better gauge room reverberation and filter out echoes."
"Ever hear (no pun intended) of audio watermarking?" asks VentureBeat. It's the process of adding distinctive sound patterns identifiable to PCs, and it's a major way web video hosts, set-top boxes, and media players spot copyrighted tracks. But watermarking schemes aren't particularly reliable in noisy environments, like when the audio in question is broadcasted over a loudspeaker. The resulting noise and interference -- referred to in academic literature as the "second-screen" problem -- severely distorts watermarks, and introduces delays that detectors often struggle to reconcile. Researchers at Amazon, though, believe they've pioneered a novel workaround, which they describe in a paper newly published on the preprint server Arxiv ("Audio Watermarking over the Air with Modulated Self-Correlation") and an accompanying blog post. The team claims their method -- which they'll detail at the International Conference on Acoustics, Speech, and Signal Processing in May -- can detect watermarks added to about two seconds of audio with "almost perfect accuracy," even when the distance between the speaker and detector is greater than 20 feet...
So how's it work? As Tai explains, the model employs a "spread-spectrum" technique in which watermark energy is spread across time and frequency, rendering it inaudible to human ears while robustifying it against postprocessing (like compression). And it generates watermarks from noise blocks of a fixed duration, each of which introduces its own distinct pattern to selected frequency components in the host audio signal. Conventional detectors would compare the resulting sequence of noise blocks -- the decoding key -- with a reference copy. But Tai and colleagues take a different approach: Their algorithm embeds the noise pattern in the audio signal multiple times and compares it to itself. Because said signal passes through the same acoustic environment, Tai explains, instances of the pattern are distorted in similar ways, enabling them to be compared directly. "The detector takes advantage of the distortion due to the acoustic channel, rather than combatting it," he added.
"Audio content that Alexa plays -- music, audiobooks, podcasts, radio broadcasts, movies -- could be watermarked on the fly," explains Amazon's blog post. It argues that this could be useful "so that Alexa-enabled devices can better gauge room reverberation and filter out echoes."
Obtain two distinct copies of the audio, diff them. Anything not common to both copies is either watermarking, noise, or compression artifacts -- and you want none of that.
The creatures outside looked from Alt-Right to Antifa; but already it was impossible to say which was which.
I read the headline and thought it was about making Alexa work nicely in an environment where it plays loud music.
Apple's HomePod does that very nicely. Instead of adding a watermark, it compares the signal entering its microphones with the signal leaving the speakers, so if you have loud music playing through your HomePod, it can eliminate that music almost completely before it starts speech recognition.
Next, if some person in the room says "Hey, Siri", it analyses the voice of the person saying the words, and eliminates what anyone else in the room is saying. Apple published a paper about this, and has some demos somewhere. One is very loud music in a room with many people talking. Phase 1 eliminates music, leaving many people talking and a bit of white noise. Phase 2 eliminates the voices of anyone except the person saying "Hey, Siri" and what's left is one perfectly recognisable voice, plus a bit more white noise. So "Hey Siri" works with loud music as long as it is played by the HomePod, and lots of people talking. What Amazon is planning here, on the other hand, doesn't seem to be something that any of the customers buying Alexa is asking for.
Your claim, precisely as stated, appears to be true but, per your link, that doesn't mean that the watermarking hasn't been broken in other ways. In fact, citation 16 regarding DVD-Ranger CinEx appears to do precisely that: detect the signal and then remove it.
The Amazon technique sounds like exactly the same crap that you get from a lot of machine-learning researchers doing security work: they don't think about an adaptive adversary. There's an entire field of adversarial machine learning that works by training a machine-learning system on the inputs and outputs of another: if you can train a neural network to insert and recognise these watermarks, can you train another one to recognise and remove them? If you haven't even tried that, it's likely that an attacker will be able to.
I am TheRaven on Soylent News