Now that Congress has declared it's not going to be paying NASA to research climate change, or think about climate change, or let anyone say the words "climate change" anymore (we're saved!), presumably there's a whole herd of NASA engineers with mortgage payments looking for new career opportunities. (One can't help be reminded of the time the Indiana legislature voted to change the value of pi to 3. Circles stubbornly continued to use the old value. Perhaps there's some kind of lesson here. Let's move on before we figure out what that might be.)
The good news. unemployed engineers, is the private sector needs your mad DSP skills to construct the Bon Scott Device.
What is the BSD, you ask?
That is a reasonable question.
We have a song. The song features a singer. We would like the song to feature a different singer. However, we don't have the cash nor the industry connections to get the new singer and a couple of booth guys into the studio and lay down a substitute vocal track. Also the singer of interest might be, well, dead. Not meant as a criticism, but nonetheless true. This is where the BSD comes in. We enter the song into the BSD, select a new singer from the pull-down menu, and, Viola! New vocals.
The name of the app is because I got to wondering what the AC/DC discography from Back in Black forward would sound like with Bon Scott instead of Brian Johnson. I don't have anything against Brian Johnson. To be honest, there's nothing specific to AC/DC about the Bon Scott device, really. You could, for example, replace Robert Plant with Bon Scott, or Katy Perry with Bon Scott, or Bing Crosby with Bon Scott. The possibilities truly are endless.
The technical requirements are straightforward:
1) A digital sample of the original vocalist
2) A sample of the replacement vocalist
3) An algorithm that turns 1 into 2
Of these, #3 is the sticky bit. One possibility is spectral mapping. You would have to comb through the original and replacement singer's discography and find common lyrics, then find coefficients that maps the spectral content of one onto the other. A convenient vehicle is cover songs, which provide a rich source of overlap. For example, compare the spectral content of Joan Jett's cover of Dirty Deeds Done Dirt Cheap to Bon's original recording:
I didn't have samples of Bon and Joan to work with, so I approximated these using my Mac's text-to-speech synthesizer, which I instructed to say "dirty deeds" using Bruce and Victoria, respectively, which I ran though spectrogram in Matlab and plotted the results. Note the subtle differences in the frequency content. We seek a transform matrix (T) such that T ⋅ S(source) = S(target), where S(source) and S(target) are the source and target spectrograms (or, alternatively, inv(T) ⋅ S(source) ⋅ T = S(target), if you want to preserve the eigenvalues, which is something people do I guess). Whether these are Bruce and Victoria or Joan Jett and Bon Scott, the concept remains the same.
Once we have the spectral transform matrix, we can apply it to the isolated lyric track of any song in the source discography of interest and superimpose the results over a vocal-less recording of the song (obtainable from the karaoke industry). Boom! Your song has been Bon Scotted, as the cool kids will begin to say (or so we hope -- it helps build brand recognition).
Continuing the current example, we can now have Joan Jett sing any song in the Bon Scott discography. From there we can quickly expand the vocal library using other covers. From Joan Jett we get Paul Westerberg (from her cover of Androgynous) and Amanda Palmer (from their common cover of the Rocky Horror theme song). From Paul Westerberg we get Paul Stanley (from the Replacement's cover of Black Diamond); from Amanda Palmer we get Ozzy Osboune (from the Dolls cover of War Pigs). From Ozzy, we get Pat Boone. And so on. Artists who do not share a cover song present a more labor intensive challenge. For this, the discography of each artist must be scoured for common words and phrases which can be used to construct the STM piecemeal. However, the likelihood of truly unconnected singers is rather low, given the incestuous nature of the music biz. Kevin Bacon effect, yadda, yadda.
On a closing note, we must acknowledge the BSD includes the potential for mischief. For example, you might change your voice into the current president and instruct the government to give all of the moneys to LabKitty. You might change into your boss and leave your coworkers voicemail instructing them to come into work tomorrow dressed as their favorite dinosaur. You might produce a version of Squealer using Justin Bieber.
I say these are acceptable risks. From sliced bread to derivatives trading, any technology includes opportunities for the debased and depraved to twist wholesome engineering goodness into a dark spectre of abuse for personal gain. The Luddites who would silence the BSD also opposed the MIRV, the automobile, and the button hole. We cannot let ourselves be turned away from progress by these shrinking violets, who would deny us a world in which we know the joy of White Christmas sung by Amy Winehouse or a version of the Queen of the Night aria featuring Axl Rose. It would do a disservice to the memory of the Scientific Revolution.
Remember, with a great app comes great responsibility. And money. Lots and lots of money.
No comments:
Post a Comment