With the release of my latest mashup, One Night Stand, I thought I’d write a blog post about how I make them. Mashup making is mostly a function of musical ability, not technical, and you can do it using entirely free software.1
You can also check out all my mashups on YouTube.
Step 0: “Hmm, that’s funny!”
The zeroth step in mashup making is just to find your source material. For me, this usually happens just by listening to music, and I feel like that’s probably going to give better results than actively looking for songs that sound alike.
My mind has gotten very good at spotting potential matches, so I’ve started recording them in a file on my computer. I currently have about fifty of these “mashup seeds”. Most of them will probably never get made; the time requirement alone ensures that. But there are a number of other reasons why a seed might not grow:
- Two songs’ keys are too far apart.
- Too hard to separate vocals from instrumentals.
- Someone else has already done it.
It’s my feeling that if someone else has already thought of a pairing, you shouldn’t do it again, even if you think you could do a better job.2 I’ve also decided that I won’t reuse a song I’ve already mashed up with something else; I’ve seen some artists that do this on YouTube and it really makes their work stand out less.
Once you’ve found two songs that work together (and do a cursory check to see if someone else had the idea first), you’re ready to start.
Step 1: Downloads
You can make a mashup with the actual retail versions of a song, but it’s better if you can find the vocal and instrumental tracks separately. Your best luck comes when the artist actually releases these tracks; searching for the name of the song plus “vocal track” or “a cappella”, or “karaoke” or “instrumental”, can sometimes turn up exactly what you’re looking for. (Watch out for a cappella and instrumental covers of the song, though.)
In particular, ACAPELLAS4U has a good collection of vocal tracks for English songs, and many K-pop singles have accompanying instrumental tracks.
Made-for-karaoke instrumental tracks should be used only as a last resort; they make the mashup sound amateur. Try eliminating the vocals from the retail version first (see step 3).
If you can’t get MP3s or AACs (m4a) for the songs, there’s a tool called PwnYouTube that might help…
Step 2: Key and Tempo
The next step is to find the two pieces that describe the structure of a song: its key and its tempo.
The simplest way to find the key of a song might be to find sheet music or guitar tabs, and simply figure out from the chords (or the printed key signature) what the key is. The trouble is that guitar tabs don’t actually tell you what the key is—you have to figure it out yourself.
Instead, what I do is pop open a virtual piano and try to play along with the song. (I used to use GarageBand for this, but recently switched to the more lightweight vmpk, with SimpleSynth as the accompanying MIDI synthesizer for vmpk’s piano keys. FluidSynth seems to be another such synthesizer that works on other platforms.) Once you’ve figured out the notes that work, you should be able to figure out the key signature…or at least find “do”, the one note that sounds good if you just keep playing it for the entire song (or almost all of the song).
It’s worth noting that if some of your songs are in minor keys and some are in major (these sound “sad” and “happy” to Western listeners), you should be consistent about whether you’re looking for the root of the major key (“do”) or the minor key (“la”). Since you’re just going to use these to put everything in the same key, it doesn’t matter which you pick, but you have to be consistent.
Failing all this, you can always cheat and try changing the pitch of one song to try and match another—then they’re in the same key.
After all that, tempo is easy. Just open up this BPM counter, play the song, and tap the spacebar on every beat. Try to be as regular as possible. After thirty seconds or so, the number should have settled down; that’s your tempo.
Step 3: Key and Tempo Matching
Now that you have your songs and their structure, it’s time to make them work together. Pick a key and tempo that’s a good average of all the keys you’re working with, then adjust every song to use that key and tempo. I use Audacity to do this, specifically the “Change Pitch” and “Change Tempo” effects. Both of them have “from” and “to” fields, so it should be a snap.
A note on pitch changes: if you’re changing by more than two semitones, the voices are going to start sounding silly, and if you change by more than three or four semitones, you’re probably better off just dropping that song (or forgetting the mashup altogether if it’s a two-song mix). I consider these “cover candidates”, where you could easily sing the two songs together in the same key, but changing the pitch of the studio recordings just isn’t going to work. I think “One Night Stand” only has three-semitone changes at most, while “The Reception” had more…and the difference is clear. For “One Night Stand”, I ended up tossing out three songs that just didn’t work.
Vocal Removal and Vocal Isolation
If you didn’t manage to find instrumental or vocal versions of a song, now’s the time to try making them yourself. Audacity comes with a plugin effect called “Vocal Remover” which can do instrumentals; I’ve also had success with the (free) third-party bx_solo and Kn0ck0ut plugins3, which can do both vocal removal and vocal isolation.
There are two principles behind vocal removal and vocal isolation, and it helps to understand them when you’re trying to make your own vocal and instrumental tracks:
Studio recordings of bands often emulate the setup of real bands, with the solo vocalist in the middle and the instruments panned to either side, or spread on both sides. There’s no “center track” in a music file, but you can compare the left track and the right track and see what shows up in both tracks. (The magic words are “Fourier transform” but you don’t have to understand how it works to use it.) If you take that, you should get an approximation of the vocal track.
The instrumental track is even easier: essentially, you’re just “subtracting” one track from the other. If the vocals are panned dead center, you can do this manually by splitting the left and right channels of a recording into mono tracks and Invert-ing the waveform of one of them. Unfortunately, this manual approach only works on very few recordings.
This whole idea doesn’t work if the song doesn’t follow the convention, does funny things with panning, etc. Backup vocals usually don’t affect vocal isolation but they make a simple vocal removal rather useless. And of course, this sort of vocal isolation only works if the recording is in stereo to begin with.
The human voice only produces sound in a certain range of frequencies: usually no lower than 100Hz for an adult man, 200Hz for an adult woman. (110Hz is the A below the C below middle C, a rare note for a non-a-cappella song.) The upper range is a little more hazy because the voice has harmonics above the pitch you’re singing; this is what gives a voice its unique quality.4
The point of all this is that by limiting the frequencies in a recording to a certain range, you can chop out instruments that are significantly higher- or lower-pitched than a human voice. This usually corresponds to high whistles, flutes, and violins, and to drums and occasionally basses, respectively. (Drums in particular are annoying; the less rhythm you have in your vocal tracks, the more freedom you’ll have later.) Audacity’s “Vocal Removal” effect can do this too, but a generic Low Pass filter can take out the high frequencies just fine (and likewise for High Pass / low frequencies).
However, if you set your lower cutoff too high, you’ll end up with a sort of old-timey radio sound, not live music. A too-low upper cutoff sounds like it’s underwater.5
You can try multiple plugins and strategies on a recorded track and see what works best. You can even do tricky things like “subtract” a found instrumental track from the original song, though this’ll only work if the instrumental is made from the same studio recording as the final song mix. But sometimes at the end of the day, you just can’t get a clean instrumental/vocal track, and you’ll just have to accept that whenever you use these vocals in your mix, you’ll get the instrumentals along for the ride. This doesn’t mean the mix is doomed to failure; I failed to get a vocal track for “Shock” or an instrumental track for “Come Back”, but “One Night Stand” came out fine.
Once you’ve finished your first song and are working on your second (or third, or fourth), it’s good to double-check your work by lining up the choruses and listening to it. This is going to sound cacophonous, cause you’re playing two songs together, but if you line the tracks up very carefully, you should be able to tell that the two songs are in the same key, and if you need to make one song faster or slower.
Here’s that “proof-of-concept” for “Tonight” against “Shock”. For the purposes of example I’ve panned the two songs to the left and right.
Hint: if the songs start together but then song B starts coming in early, song B is too fast. If song B feels early the whole time, move it forward until the starts match up, and try again.
Step 4: Cut, Splice, Fade!
Now that you’ve got all the ingredients, it’s time to start making your mashup! I used to use GarageBand for this, but I’ve recently switched to Soundtrack Pro, which has a slightly nicer interface for dealing with prerecorded audio (you can disable snapping, you can overlap pieces of a song and have them auto-fade into one another).
I tried using Audacity once, and it worked okay; the problem is that all of Audacity’s edits are destructive. If you cut a clip short, you can’t drag it out to make it longer again. If you add a fade in or fade out, you can’t undo it later. And it’s very hard to smoothly lower the volume partway in the middle of a clip. So if you’re not on a Mac, you’ll have to find another program to work with here, one that allows nondestructive editing.
In order to help keep me sane, I started color-coding my tracks on larger projects, using green for “regular”, orange for “instrumental”, and blue for “vocals” (and gold for “sound effect”). Another good scheme would be to set different colors for different songs.
In the screenshot, you can see a manual volume adjustment from the regular version of “Shock” to the instrumental. Whenever you fade from one track to another, you want to fade in before you fade out. This will get closer to the right perceptual loudness (remember, 10 decibels more means 10 times as loud).6
Actually, that’s pretty much the only mashup hint I have to share. Fade in before you fade out. Everything else is a matter of style. Oh, and don’t be afraid to move vocals around to new rhythms; it can work really well sometimes.
I tend to work straight-through from start to finish, with several passes over what I’ve done fixing things. This has the advantage that it’s often really hard to move things around in most audio programs once you have them set up the way you want.
Finally, when I think I’m done, I try to make myself leave the computer and come back the next day. Chances are I’ll notice something else I need to fix.
Step 5: Levels
When you’re putting your mix together, chances are you’ve only paid attention to how it sounds locally. If you’re not careful, and especially if you have lots of tracks, it’s possible that the end of the song is much louder or much quieter than the beginning. Now’s the time to listen to the song all the way through and see if you notice any weird jumps. You can also try jumping around in the song and see if anything is much louder or much quieter.
I suggest doing all this with headphones on and the volume turned up so that the song is just barely not uncomfortably loud. That way, if there’s ever a spike (too loud is more common than too quiet in my experience), you’ll notice it.
Then, because you can’t trust your own ear, you should listen through the song again, this time watching the level meter for the final mix. Pick a level you absolutely will not go above and make sure everything in your song stays below it.
If you do have problems, note that they’re often in the percussion (a sudden drum hit can be much louder than a guitar), so it’s better to look at your instrumental tracks than your vocals. You could even set an equalizer on that part of the track that reduces low-frequency sounds.
Also note that in general, verses are quieter than choruses. You can do this deliberately or just make everything the same, but your verses should almost never be louder than your choruses. (Unless you have that kind of “calm, quiet, contrasting” chorus, but even then don’t overdo it.)
If you have any songs you included without changing the pitch at all, this is also a good time to sneak in a pitchbend of maybe 50 cents (half a semitone). This helps slow down automated copyright violation filters *cough cough* and won’t usually change the feel of your track that much.
Step 6: Album Art / Music Video
If you’re just distributing your music on your own personal site, or on Soundcloud, you can get away with skipping this. But the most common distribution venue is YouTube, where people want to have something to look at while they listen.
For most of my mashups, I just make a visual combination of the artwork associated with each of the songs. You can use any photo editor for this: Photoshop or GIMP or even MS Paint. I personally use Acorn for most of my simple tasks these days, and switch to GIMP if I want to do something that my old version of Acorn can’t handle. You can even use Keynote or PowerPoint if you’re just going to collage several things together.
Unfortunately, I haven’t found a great way to put artwork and music together in a single movie file. I’ve been using my old copy of QuickTime Pro 7 to “Add to Selection & Scale” the image into the song file, but there ought to be an easier way.
Sometimes, though, if every song in the mashup has a music video, I’ll make a composite video for the new mashup. I’ve done this for “On the Right Boat”, “The Reception”, and of course my latest one, “One Night Stand”.7 For this you can use any video editor, from Windows Movie Maker to iMovie to Final Cut Pro. Ideally, though, you want an editor that supports multiple video tracks and makes it easy to cut up your video.
When using Final Cut Pro, I lock the audio tracks from the start, so I don’t accidentally mess with the mashup. Then I make it so that adding a video doesn’t include its associated audio.
The one thing to be careful of in making a video is that if any video clips have people singing, they should match up with the words in your mashup. So, if you faded this part of the song to instrumental, you should ideally put another video clip in that’s just a scene (without mouth movement), or even a clip from another song.
Actually, this is one of the best things about editing the movie separately from the song—your video doesn’t have to match the audio, and usually shouldn’t. Things stay more interesting if you “reference” the other songs in your mashup, even when they’re not active in the audio.
Like editing the regular mashup, I usually try to step away from the computer when I finish the video and come back later, where I’ll then notice something else I need to fix.
Step 7: Upload to YouTube
…and ignore the copyright warning, if one comes up. Mashups are arguably “fair use” under US copyright law, but I say “arguably” because they are clearly “derivative works”. It’s also unclear whether they’ll affect the market value of the song positively or negatively (anecdotally, I’ve seen both), and while they’re “noncommercial” they’re definitely not “educational”. (Selling your mashup is almost certainly illegal without permission.)
Then tell people about it on Facebook. Especially me.
Step 8: Buy the Songs
It may just be me, but I think once you’ve made a mashup, the least you can do is buy the original songs yourself if you haven’t already. You’ve just made use of the artist’s voice and their instrumentals; the least you can do is show them a little love on Amazon MP3 or on iTunes.
And that’s how I do it. I got started almost by accident with “On the Right Boat”, which was a Step 0 that stuck in my head particularly well. Since then, it’s been a fun on-and-off free time activity, and I’m proud of finding these patterns and putting them together to make something new. It’s a pinch of musical ability with an ounce of patience and persistence.
Congratulations! You’ve made a mashup!
…Even though I don’t. ↩︎
…unless they did a really bad job, in which case you still cite them as inspiration. Or unless you’re adding additional songs, but even then it’s probably better to just stay away from one of the original two. ↩︎
Kn0ck0ut is unfortunately a dead codebase and hasn’t been updated in a long time; the online plugin download is Windows-only. However, I managed to get it kind-of-working on Mac OS X a while back, so if you want me to send you a copy I can. There’s also a recent port to the LV2 audio plugin format, but Audacity doesn’t support LV2 yet! ↩︎
Actually, the different harmonics are how we tell vowels apart; a high “first formant” indicates a vowel where the mouth is more open, like [ɑ] (“ah”) as opposed to [i] (“ee”). ↩︎
Transmitting lower frequencies on a radio, at least AM radio, requires a larger antenna. (I’m not sure if the same is true for FM radio or not.) Water is denser than air so higher frequencies are usually damped out of existence by the time it reaches our ears. SCIENCE! ↩︎
Really, the correct fade is a curve, which Soundtrack will do for you if you overlap two tracks. But that doesn’t help if you want to only fade out partway. ↩︎
As the last two show, you can even make this work if most of the songs have MVs. For the one or two that don’t, you’ll have to find another video from the same artist and figure out how to chop it into pieces, then reconstruct it to fit the song you used—bonus points if you can match up mouth movements on some of the lyrics. This worked well for “Stay” (Jay Sean) in “The Reception”, not so well for “Because I Love You” (Groove Coverage) in “One Night Stand”. ↩︎