Back to 2020 gallery

Effects of Ghosted Transients on Information Retrieval in Music

Matthew Bradford

In the early 1990s, a particular style of hip-hop music called Chopped and Screwed was developed in Houston, TX. The style is characterized by playing the same song one beat out of sync on two turntables and moving the crossfader rapidly between the two tracks to make 'chops' in the track that extend and contract the music nearby it in time perceptually. The chopped track is then recorded on a tape deck at a much slower speed, which stretches the track linearly in time. The overall effect is one of space and gravity.

What I am testing in this illusion is the effect on rhythm perception of adding gaps around transient peaks in the audio signal. These transients are usually associated with the drums of a particular song and are therefore linked to the rest of the elements of the track as the drums keep time for everything else. What I have found is that the transient coding exposes a non-linearity in music perception. As we lower the sampling rate, the missing transient information from previous tracks helps the listener to refactor the time signature of what they know to be music. The unchopped versions that are slowed down are more vague at lower speeds than their chopped equivalents.

The stimuli were chopped in Ableton Live and then slowed down using wav_write, a module of scipy. The song is the beginning of "Wait Until Dark" by Henry Mancini with Sue Raney. Wait Until Dark is a very good movie with Audrey Hepburn and Alan Alda.

Unstretched:

Chopped Clean

90_Speed

Chopped Clean

80_Speed

Chopped Clean

70_Speed

Chopped Clean

60_Speed

Chopped Clean

50_Speed

Chopped Clean

40_Speed

Chopped Clean

30_Speed

Chopped Clean

20_Speed

Chopped Clean

10_Speed

Chopped Clean
RIP Screw

Comments

Malinda McPherson

a) I listened a few times and found that it was easier/faster for me to start tapping along to the 'chopped' versions compared to the clean versions. I definitely think that the effect increases for the slower speeds, but then around 30 both rhythms are hard to follow. I also don't really hear the gaps all that clearly until it slows down to about 60% - It sounds like something is turning off and on at higher speeds, but I wouldn't be able to tell you for certain just from listening that the sound fully stops.
b) I think it's a really cool set of demos! I agree that it works, but to me it is a moderate effect in mainly the 40-60 slow down range. I'm curious about what you mean by this exposing a non-linearity?

Matthew Bradford

@Malinda McPherson,
There may be a sampling resolution breakdown around 30% speed or so. The maximum sampling rate I could start with (which I used) was 32bit 192kHz. To skip ahead to your second comment, what I mean by a non-linearity is that the effect does not linearly increase starting from when the speed is first slowed down. The jump in effect from 50% to 40% is larger than the jump between 90% and 80%.

In order to improve this illusion, it would be prudent to manage the sampling rate more intelligently. The relationship between original tempo and playback speed has a periodic component, so perhaps examining ratios between original sample rate and new sample rate multiplied by some tempo factor would be a better analogue for examining this effect. Depending on the shape of this curve, I would then target the illusion demos more precisely in the 40-60% speed range.

Anonymous

@Matthew Bradford,
Thanks for the clarification!

Richard McWalter

a) The music clip is slowed down and transients are included were there are drum beats. The beat of the "chopped" versions are pretty easy to follow for the slower speeds (~50) but the beat for the "clean" versions are hard to follow the same slow speeds.
b) I think it's a nice investigation into a perceptual phenomenon related to music. I agree that the transients help to identify the time signature (or beat) or the music for the more time-stretched samples. This is particularly the case for the 50 and 40 speed samples. It's quite easy to track the beat with the transients, but for the clean reference the beat becomes more ambiguous. For each of the samples, I found that the effect was most prominent during the last half of the keyboard section.

Nicholas Guiliano

(not assigned, not to be graded lol) This is really cool. I feel like your observation is correct.

Matt

@Nicholas Guiliano,
Thanks so much!