Sounds of Silence

Tuesday, February 1, 2011
By Sharad Goel

When I was a kid we had a small black & white television that never quite worked right. If you positioned the rabbit ears just so—and no one turned on the kitchen blender—the signal to snow ratio would be just barely high enough to make it through an episode of “The Cosby Show.” Humans are great at filling in missing visual information. Talis Bachmann, for example, demonstrated that people could accurately identify faces from a coarse grid of only 16 x 24 pixels, comparable to the left-most image of Einstein below.

The subjects filled in the missing music in familiar songs, hearing it in their mind’s ear.

So how well can people recognize degraded audio signals? I started thinking about that question when reading Musicophilia,  fascinating “tales of music and the brain” by Oliver Sacks. In it Sacks relates the story of a friend who sits down to listen to his favorite Mozart record and proceeds to thoroughly enjoy the music, only later realizing that he had never actually turned on his record player. It seems likely that either Sacks or his friend is indulging in poetic hyperbole. Nevertheless, “musical imagery”—the subjective experience of hearing music absent auditory stimulation—occurs quite regularly in most people and is a well-documented physiological phenomenon. For example, Kraemer et al. presented people with familiar and unknown songs in which short sections were replaced by gaps of silence. Using functional MRI, they demonstrate that the silent gaps embedded in familiar songs induced greater activation in the auditory cortex than did the gaps in unknown songs. The subjects, in other words, filled in the missing music in familiar songs, hearing it in their mind’s ear.[1]

Now on to this week’s puzzle. Each of the mystery songs below was sectioned into super short segments—one-tenth of a second long—and 90% of these segments were replaced with silence. Consequently, for every second of play time, you hear one-tenth of a second of music. How good is your mind’s ear at reconstructing the songs? Bragging rights go to the first person to name all five songs correctly. Leave us a comment with your guesses. Good luck!

1. Mystery Song #1

2. Mystery Song #2

3. Mystery Song #3

4. Mystery Song #4

5. Mystery Song #5

NB: Einstein was pixelated with Processing, and the audio clips were generated with SoX. Thanks to Walter Kim and Jake Hofman for pointing me to those cool tools.

Illustration by Kelly Savage

Addendum

Congratulations to puzzle winner Kevin Canini, who successfully identified all five mystery songs less than hour after they were posted! Honorable mention goes to Jeff Ely for identifying the super tough challenge song, which I actually thought no one was going to figure out. If you didn’t identify all the tunes and want to keep playing, here are progressively less degraded versions of the mystery songs above. The final sample in each sequence is the original, undistorted version.

1. 90% Degraded » 80% Degraded » 60% Degraded » Original Clip

2. 90% Degraded » 80% Degraded » 60% Degraded » Original Clip

3. 90% Degraded » 80% Degraded » 60% Degraded » Original Clip

4. 90% Degraded » 80% Degraded » 60% Degraded » Original Clip

5. 90% Degraded » 80% Degraded » 60% Degraded » Original Clip

Footnotes

[1] Sacks makes a much stronger—and ostensibly false—statement when describing Kraemer et al.’s work, claiming that “the silent gaps in familiar songs were not consciously noticed by their subjects.” While participants may have heard the missing music in their mind’s ear, they almost certainly knew it was imagined. To wit, the gaps in even very mildly degraded versions of the mystery songs are quite jarring.

Tags: , ,

  • http://www.cs.berkeley.edu/~kevin/ Kevin Canini

    That was surprisingly easy!

    1. Bad Romance – Lady Gaga
    2. Satisfaction – Rolling Stones
    3. Hey Jude – Beatles
    4. Ode to Joy – Beethoven
    5. Respect – Aretha

  • http://cheeptalk.wordpress.com jeff ely

    1. hey jude
    2. ode to joy
    3?
    4. satisfaction (detectable from the first chunk alone)
    5?

  • andy

    1. hey jude – beatles
    2. ???
    3. respect – aretha franklin
    4. satisfaction – rolling stones
    5. bad romance – lady gaga

  • http://hampiesandwiches.blogspot.com eileen

    I did one listenthrough. The only one I actually got was Aretha, although I could figure out bits of other stuff (meaning genres, i.e. Ode to Joy was obviously a large orchestral choral piece. Also I thought the Beatles sounded like Rivers Cuomo, which is unsurprising).

    One thing I found interesting: even though I told myself I was trying to figure out what the songs were, my brain was all like “let’s see if we can use our musical knowledge to predict what’s coming next instead!”

  • http://messymatters.com/sharad Sharad Goel

    Kevin: Wow, nice job! Are you up for a bonus challenge? Mystery Song #6 is 95% silence: you hear one-tenth of a second of music for every two seconds of play time. Can you name that tune?

  • http://messymatters.com/sharad Sharad Goel

    Just to clarify, Kevin’s answers are all correct, but his numbering is a bit different from how the songs are labeled in the post.

  • http://www.consultingstatistics.org Basil

    I didn’t think it was that easy. I had my class help me.
    We came up with:
    1. Hey Jude – Beetles
    2. ?Sounded like a choir?
    3. Aretha Franklin – Respect
    4. ?We thought of an old rock song?
    5. Bad Romance – Lady GaGa

  • John Cox

    This reminds me of sine-wave speech distortions.

    The linked page has a table of audio snippets. Recordings in the right column contain a sentence of speech, one per row. The middle column contains that same sentence, distorted. Can you tell what the sentence is for each row without listening to the recording on the right?

    If not, listen to the original and then listen to the distorted recording in the middle again. With this new information you’ll make out the distorted recordings easily. I also found that after cheating on the first two sentences I was able to identify the rest without listening to the original. My brain rapidly got better at teasing speech out of this particular deformation.

    The amount of processing your brain does to ‘raw’ sensory input before you consciously perceive it is really surprising.

    (Also, I wonder how possible it would be to identify the Bach fugue at the beginning of Bad Romance if you replaced 90% of it with silence. It’s already hard to recognize because it’s played so fast.)

  • http://dreev.es dreeves

    @John Cox, are you saying you can identify the nursery rhyme on that page? I couldn’t, even after training on all the example sentences.

    This reminds me of another interesting phenomenon: priming. If you made up some snippet of speech that matched the rhythm in that example I would think I was hearing it plain as day. Here’s a funny example: http://languagelog.ldc.upenn.edu/nll/?p=1944

  • http://cheeptalk.wordpress.com jeff ely

    mystery song #6 is I Feel Good

  • http://cheeptalk.wordpress.com jeff ely

    And I should have gotten Aretha Franklin but my excuse on Lady Gaga is that I have never heard that song.

  • http://messymatters.com/sharad Sharad Goel

    @jeff, right on—very impressive!

  • Mohammad

    there are two big differences between image and music: first, image is 2 dimensional but music is 1 dimensional. so after removing a fraction of the data points, the 2 dimensional image is “more connected” than the 1 dimensional song. second, you have random access to an image when you look at it, but you experience music as a stream.