Investigation of Scaler audio and MIDI detection Part I

I have been performing some tests on MIDI (and related audio) chord detection in Scaler, and comparing this with other means of detection.

These preliminary exercises involved {a} Scaler, for both MIDI and audio, {b} Band in a Box, for MIDI, {c} Cakewalk, for MIDI, and, {d} deCoda, for audio.

The outcome was that scaler had by far the most technically accurate MIDI detection, but that it had a number of ‘quirks’, and in some cases one or other of the other tools had better results. Scaler was less good with various classes of audio tests.
(To download the files referred to, go to . | BTinternot and download detection-a )

As a poor theory amateur, it seems to me that the minimum simultaneous notes needed to usefully detect a chord is 4, namely (i) the root, to detect the key (ii) the third, to detect minor tonality, (ii) the seventh to detect dominant tonality, and, (iv) any upper extensions, such at 9th etc.

Test 1 : pad chords, legato, monotimbral

The test sequence is 4 bars, notionally G/ Am / C / F, but made up of (bar 1) G1/D2/A2/ D3 (bar 2) A1/E2/G2/ E3 (bar 3) C1/G1/E2/G2/C3, ad (bar 4) F1/C2/F2/A2/C3/F3. Note that the diatonic sequence in triads would be G B D / A C E / C E G / F A C . (see ‘nk gamcf.jpg’

Scaler detected this as Gsus2 / A7(no3) / Cmaj / Fmaj (‘nk gamcf scaler.jpg’). This is correct, but the scale it proposed was G Mixolydian. This is interesting, as although the detection was theoretically correct, a guitarist would have probably played the flat 3rd (by Am pentatonic notes there) and not the flat 7, where they would normally have emphasised over a dominant 7th chord. Listening to the chord, however, implies a minor tonality rather than a Mixolydian one, and pentatonic fits better.

However, if this was rendered to audio in Live, using a pad from Omnisphere (‘nk gamfc audio.wav’) Scaler detected this as 5 bars D5/Dmaj/Cmaj7/C5/Fmaj, and declared it to be in D Mixolydian. (see ‘nk gamfc audio.jpg’). This was strange, and nowhere near where it should be (and why 5 bars ?)

In BIAB, the sequence was detected as Gmaj / Am7/ F6 / Cmaj7 . It was interesting to note that BIAB picked up the missing third, and played the notes that one would have done by ear.

However, the trigger for further investigation was that deCoda served up G/Am/C/F, and arguably was closer to the actual tonality than MIDI detection. Did this mean that precision in detecting MDI might be being bought at the cost of following notes and not tonality ? For the next episode, see Part II

1 Like

A series of lamps from the bottom…

I downloaded detection-a and I’ll test them with Broomstick Bass that seems having a fine MIDI chord detector

Broomstick Bass recognizes all 4 chords and plays the bass on
I tried the Toontrack EZbass Audio Sender, and it recognized well the audio