4 ms·

After 20 Minutes of Listening, New Adobe Tool Can Make You Say Anything

20 points by michaelmior 10y ago

rl3 10y ago
Watermarking this is virtually pointless, if not detrimental. The only utility is perhaps shielding Adobe from liability. Beyond that, it's simply delaying the day when the authenticity of uncorroborated audio recordings falls under heavy suspicion by default. In the mean time, the technology will exist in other, open-source incarnations soon enough (if it doesn't already, albeit in far less user-friendly forms). The efficacy of these in successfully framing people will be far higher if the output of the most popular tool (Adobe's) is watermarked. Intelligence agencies probably already have the technology perfected, and I'm sure the output is devoid of watermarks.
erroneousfunk 10y ago
This is fantastic. I've dabbled in audio editing (rearranging songs, terrible remix projects, lining up the beats in two tracks, adding in some cool movie quotes over the instrumental solos for fun) and it REALLY feels like photo editing in Microsoft Paint! You're just staring at these wave forms, zooming in, moving the cursor over by a pixel here and a pixel there, listening to the same half-second blip of audio over and over again, lining spikes up with other spikes juuuuust right, until your brain goes numb. Sure, some people are probably much better at this than I am, but some people are probably also much better at manipulating images in Paint than I am, and that's not a good reason to make it the standard. Honestly, I can't believe this type of commercial product hasn't been developed earlier. In retrospect, it seems so obvious and useful!
vinchuco 10y ago
Great, now blind users can listen to text in the voice of, say, Donald Duck. Or it can potentially make a whole movie sound like only one of the characters. But more seriously, this would do wonders for extortionists ¯\_(ツ)_/¯ In fact, there's already ways to generate text using your common speech patterns. How hard is it to get the inflection right? Does it work in Chinese? A little video editing here and there and some control of the media and you just made that politician say something "undemocratic". Scenario: "Wait, didn't you call me telling me to leave the keys to the safe in the kitchen counter?" And when you make a call, who else is listening? Voicemail isn't stored in your cellphone. For a more positive outlook, perhaps it can be used to better train speech recognizes such as YouTube's , just take the diff of audio vs generated cc as training measure (?).

20 points by michaelmior 10y ago

rl3 10y ago

Watermarking this is virtually pointless, if not detrimental. The only utility is perhaps shielding Adobe from liability. Beyond that, it's simply delaying the day when the authenticity of uncorroborated audio recordings falls under heavy suspicion by default. In the mean time, the technology will exist in other, open-source incarnations soon enough (if it doesn't already, albeit in far less user-friendly forms). The efficacy of these in successfully framing people will be far higher if the output of the most popular tool (Adobe's) is watermarked. Intelligence agencies probably already have the technology perfected, and I'm sure the output is devoid of watermarks.

erroneousfunk 10y ago

This is fantastic. I've dabbled in audio editing (rearranging songs, terrible remix projects, lining up the beats in two tracks, adding in some cool movie quotes over the instrumental solos for fun) and it REALLY feels like photo editing in Microsoft Paint! You're just staring at these wave forms, zooming in, moving the cursor over by a pixel here and a pixel there, listening to the same half-second blip of audio over and over again, lining spikes up with other spikes juuuuust right, until your brain goes numb. Sure, some people are probably much better at this than I am, but some people are probably also much better at manipulating images in Paint than I am, and that's not a good reason to make it the standard. Honestly, I can't believe this type of commercial product hasn't been developed earlier. In retrospect, it seems so obvious and useful!

vinchuco 10y ago

Great, now blind users can listen to text in the voice of, say, Donald Duck. Or it can potentially make a whole movie sound like only one of the characters. But more seriously, this would do wonders for extortionists ¯\_(ツ)_/¯ In fact, there's already ways to generate text using your common speech patterns. How hard is it to get the inflection right? Does it work in Chinese? A little video editing here and there and some control of the media and you just made that politician say something "undemocratic". Scenario: "Wait, didn't you call me telling me to leave the keys to the safe in the kitchen counter?" And when you make a call, who else is listening? Voicemail isn't stored in your cellphone. For a more positive outlook, perhaps it can be used to better train speech recognizes such as YouTube's , just take the diff of audio vs generated cc as training measure (?).