We generate MIDI from audio, and look at what’s new in Live 11.1.
Live’s melody, harmony and drum audio‑to‑MIDI conversion algorithms are typically used to extract MIDI sequences from single‑part audio clips in order to either edit the parts or use different instruments to play them. You’ll find a detailed analysis of that process in the July 2013 Live column. Here we’ll look at a different use for conversion: deciphering individual parts in a multi‑part audio clip. We’ll also take a look at some of the new features in Live 11.1.
This example starts with three MIDI clips, renders their mix to a single audio clip and then uses Live’s three conversion algorithms to extract melody, harmony and drum MIDI clips to compare to the original MIDI clips. The reason for starting with MIDI clips is to make clear the differences that can crop up when trying to analyse a complex audio clip (the usual starting point) using the conversion algorithms. Conversion does work best on single‑part audio clips of the appropriate type, but as we’ll see, it can be surprisingly useful for deciphering more complex audio material.
In Screen 1, I’ve started with bass (Funky Retro Bassline C Minor 145 bpm), piano (House 90s C Minor 128 bpm) and drum (Hip Hop Swing 105 bpm) MIDI clips from Live’s Core library. Each of the these MIDI clips comes with its own Live Instrument Rack, which is imported automatically when you insert the clip on an empty MIDI track. Rendering their mix creates a single audio clip, which I’ve inserted on a new audio track. You access Live’s three conversion algorithms for a selected audio clip either from Live’s Create menu or from the clip’s or clip editor’s context menu (right‑click/Control‑click). Applying one of the conversion algorithms to the audio mix results in a MIDI clip on a new MIDI track holding a default instrument specified in your User Library/Defaults/Audio to MIDI folder. In this example, the best comparisons result from moving the MIDI‑conversion clips back to the MIDI tracks holding the original MIDI clips so that they play their intended instruments.
The before‑and‑after comparison when playing all three clips is surprisingly listenable, but when you compare them one instrument at a time, the inaccuracies stand out. The converted bass has some wrong notes, which are mostly stolen from the piano part. The converted harmony is a bit more accurate, but has missed some notes and has also stolen some from the bass. Drum conversion is always restricted to three kit pieces (kick, snare and hi‑hat) and places notes at every transient marker in the original audio track. (Transient markers are the targets in all the conversion algorithms.) One way to improve on the drum conversion is to make a copy of the source audio clip, delete each transient marker where there is no drum and use that clip for the drum conversion. (You could do that for the melody and harmony parts as well, but it’s probably not worth the effort.) None of the inaccuracies are too surprising, but they show you what to look for when trying to convert a multi‑part audio clip to MIDI clips. The bottom line is that, in spite of the shortcomings, you can get a lot of useful information.
Conversion does work best on single‑part audio clips of the appropriate type, but it can be surprisingly useful for deciphering more complex audio material.
When you convert a complex audio clip to MIDI clips using the provided algorithms, problems with melody and harmony conversions typically involve notes out of range or out of key. In Screen 1 there are no notes out of key (Eb), but notes above F1 don’t belong in the bass part and notes below G2 don’t belong in the piano part. Without the original MIDI parts, you will have to pick out these ranges by ear. Having done that, you can correct notes in the converted MIDI clips either manually or by using Pitch MIDI effects inserted in parallel chains of a MIDI Effect Rack, with their key zones set to route incoming notes to the Pitch effect for the correct octave shift. When necessary, you can also use Scale MIDI effects to keep things in key. After that, it’s up to your ear to identify and correct differences between the original and the melody and harmony conversions.
Another option is to use a filter effect to focus the source audio clip on the range of the instrument being converted. Screen 2 shows the results for the bass and piano conversions from Screen 1. The EQ’d bass conversion nailed it. The piano conversion is better, but there are a few missing notes and some wrong notes, which I’ve deactivated.
Beyond deleting transients, you’ll need to correct drum conversions by ear, but you have a lot of room to play around. Besides adding and deleting notes, you can swap kit pieces. That’s especially useful with the hi‑hat lane, which you can think of as an ‘everything else’ lane. It’s less important to mimic the source drums than to make the part sound good.
Although Live 11.1 offers no monumental changes, it makes a number of things easier. One of my favourite improvements is how you bring in devices from tracks in other sets. Previously you had to bring in the external track, or if the target track is empty, a clip from the external track. If that device doesn’t do the trick, you have to repeat the whole process. In Live 11.1, revealing a track’s contents in Live’s browser displays a Devices icon at the bottom of the list (Screen 3, top). You can drag that into the Device view of any track to replace or add to existing devices. (Type ‘.ALS’ in Live’s Search bar to restrict the browser’s content to items with devices.)
Live’s Clip view has a new look. You can arrange the Clip and Tools panels vertically or horizontally. Vertically, you scroll through them folding or unfolding each panel as needed (Screen 3, middle left). Horizontally, you get separate Clip and Tools groups with tabs to select the panel displayed in each group (Screen 3, middle right). You choose your view by click‑dragging the right edge of the panel display from folded (no panels displayed) to one panel (vertical scrolling) or two panels (horizontal display with tabs). Live’s View menu offers automatic switching from horizontal to vertical view when the clip viewer is above minimum height. (That gets disabled when you change views manually.) You can also toggle clip view between its current height and full height with Ctrl+Alt+E/Cmd+Opt+E. A welcome addition to the clip editor when multiple MIDI clips are selected is that you can now select, move, copy and paste notes from several of the clips at the same time.
In Arrangement view with a nested track display (a Group’s tracks or a track’s take or automation lanes), you can use the left and right arrow keys to navigate the lanes and tracks. Left arrow takes you from lane, to track, to folded track, and finally, to Group if the track is in a group. Right arrow unfolds the Group tracks and track lanes, and you then use the up and down arrow keys to navigate through them. You can also duplicate selected take lanes with Ctrl+D/Cmd+D.
The new Shifter audio effect is a significant upgrade to its predecessor, Frequency Shifter. ‘Frequency’ is dropped from the name because Shifter now offers pitch shifting (semitones and cents) as well as frequency shifting (Hertz) and ring modulation. It also features a more flexible LFO with optional envelope‑follower modulation and a built‑in feedback delay. The example at the bottom of Screen 3 starts with the ‘Vocal Demo Harmony’ female vocal clip from Live’s Core library. The Coarse setting, which is modulated by the separate LFO device on the left, transposes the input, and with Wide activated, the right channel is inverted and further transposed by the Spread (cents) setting. That is followed by a feedback delay. The orange and blue curves at centre reflect the pitch shifts for the left and right channels. Their effect is determined by the LFO type, Rate, Amount, Duty C, Phase and Offset settings — the possibilities are endless. The envelope follower provides further pitch modulation. The 50 percent Dry/Wet setting mixes in a bit of the input to provide a third pitch component. A quick listen to the Freq and Ring modes reveals why Pitch mode is a welcome addition.