PG Music Home Forums General Discussion Off-Topic Song Master Pro

How does the chord transcription feature compare to Chordify?

Joanne, I have not tried Chordify. Up until Song Master Pro, my go-to program for chord transcription (besides me) has been Chord AI, an iOS app. If you are familiar with that excellent app, perhaps it will help to know that Song master Pro is far better.

I just ran a song (Naufragio as sung by Ibrahim Ferrer) through the demo version.
It did a very nice, if overly complex version of the chords.
Often it would change chords entirely on a bass line change (passing tones). Mostly enharmonics, but something you notice.
Nice work, but I would simplify before before bringing to rehearsal.

Nevertheless, Good stuff overall.
If I could import the xml into biab and get decent charts that would save a lot of work.
I guess you really won't know the cleanup effort till you buy the stuff.

I think I'll give them the $$$

Note that Song Master and Song Master Pro have just been updated, to v4.0 and v2.0 respectively.

The Song Master v4.0 release notes:

https://aurallysound.com/pages/release-notes

The Song Master Pro v2.0 release notes:

https://aurallysound.com/pages/release-notes-pro

Here's the download page to save you some clicks.

https://aurallysound.com/pages/download

I export music xml from Song Master Pro > Musescore all the time and it works flawlessly. It's an incredible time saver.

Unfortunately, I've made a few testes importing those same music xml files into BIAB and the result hasn't been good. In fact, I've never been able to use Biab music xml import / export feature successfully.

Yes, that's the same experience that I have had.

I did some more tests looking into the MusicXML files.
Both BIAB and Songmaster PRO are using MusicXML 3.0
But they seem to handle the chord ”qualifier/extenstion specification" differently!

Looking in the MusicXML files for some sample chords as an example.

SongMaster PRO exports a chord E7 resp E9 as:

<root>
<root-step>E</root-step>
</root>
<kind>dominant</kind>

<root>
<root-step>B</root-step>
</root>
<kind>dominant-ninth</kind>

And BIAB exports a chord E7 resp E9 as:

<root>
<root-step>E</root-step>
</root>
<kind text="7">dominant</kind>
</harmony>

<root>
<root-step>E</root-step>
</root>
<kind text="9">dominant-ninth</kind>

(strange that it should differ if they using the same the standard).

MuseScore uses MusicXML 4.0 and seems to import both versions correctly as far as I can see.
as well as GuitarPro 8 uses MusicXML 2.0 and seems to import both versions correctly as far as I can see.

But BIAB misses the chord ”qualifier/extension specification” when importing MusicXML from SongMaster Pro ie an E7 or E9 chord comes in as E and similar for others. But the basic chord structure for the song seems to be imported ok (ie chords without qualifiers/extensions). Also BIAB seems to be able to import its own created MusicXML correctly.

Maybe there is a configuration issue somewhere (or some simple change one can modify in the MusicXML file to get it working?)

I don't know if there is one correct way to implement the standard or if both ways are equally correct. But interestingly other software companies seem to be able to handle it so it works for both cases.

Music XML in BIAB also has issues with importing songs with time signature changes.

I understand that PG Music has been asked to consider updating the product to support Music XML 4.0

On of the many complications of MusicXML in particular and XMLs in general is that so many tokens have options/attributes that one may or may not choose to use. Looking at MusicXNL 3.1, of which I happen to have a copy, I'm fairly confident that the that the text="7" attribute is optional, as are various other attributes of kind (use-symbols, stack-degrees, parentheses-degrees, bracket-degrees, print-style, haligh, valign). And that's just one token of many.

Also of course, the specification says what tokens and attributes are/mean, and in what context, but how that's interpreted by the software may be imperfect.

pretty cool program. i have demoed the thing and it does do a lot of fun things.

Too bad. I was hoping for a shortcut to manual labor.

As mentioned elsewhere, support of a current version (4.0?)of music xml by PG music would be helpful.

Maybe it does not help with v4.0 for this specific problem with importing chords.

If v3.0 allows for using an optional textstring as a chord specifier rather than the detailed <kind> chord specifier as it seems to be at the moment (see above). Then maybe this problem will remain in 4.0 unless BIAB changes the way to interpret the chord specifier (only a speculation).

I just ran a test with a Brazilian song (lots of complex chords). While Song Master Pro (updated) read the chords well, the Music XML loaded into BIAB showed only the chord root. I had to type over every chord to include 7, m7, Maj7, 9, 13, etc. It did get the roots in the correct positions in measures and surprisingly did read slash roots.

I need to read the code to find out where the error is.

features that I’d like with chord AI

it can interact with YouTube for it’s source material

Is available on iPad

I have done some more testing on importing MusicXML generated by SongMaster Pro into BIAB.
As mentioned in posts above when importing a MusicXML file that has been generated by SongMaster Pro directly into BIAB it only loads root chords no extensions.
But maybe there is a workaround.

If I import the MusicXML file into MuseScore it loads all chords correctly. If I then export the same score (the MusicXML just loaded) out again as MusicXML from MuseScore and then load it into BIAB it may work. But it did NOT!

If I copy the same score (the MusicXML file I just loaded into MuseScore) into a fresh new empty score in MuseScore and from there export it out as MusicXML from MuseScore and then load it into BIAB….

… It seems to Work FINE.. (all chord extensions are there).

It seems that MuseScore4:
1) exports scores that have been created from scratch in MuseScore correctly in a format that BIAB can read (ie it includes both the ”text string chord extension names” as well as the <kind> chord extension definition parameter).
2) but when exporting a score that has been loaded from MusicXML it seems just to just export exactly the same MusicXML file that was originally loaded.

I don’t know what the standard states or which of the software that need change or maybe both (but MuseScore seems to import MusicXML files correctly both from SongMaster Pro and BIAB)

So there might be a workaround a bit heavy but it seems to work.

EDIT ------------
To me it looks like that the chord extention is specified by:
1) the <kind> chord extension parameter in MusicXML but also
2) an optional ”name text string” parameter (it is only my guess that it is optional I don’t know if it is mandatory)

SongMaster Pro only adds the 1) the <kind> chord extension parameter and not the 2) ”name text string” parameter.

But this seems to work fine with MuseScore (as well as GuitarPro8 and Notion 6) and it loads correctly with the correct chord extensions These programs do not seem to need the 2) ”name text string” parameter, they work fine with only the 1) the <kind> chord extension parameter.

But this is not the case with BIAB it seems not work without the the 2) ”name text string” parameter. it seems that BIAB requires the 2) ”name text string” parameter in order to interpret the chord extension and BIAB seems to ignore the 1) the <kind> chord extension parameter if it is given alone without the 2) ”name text string” parameter.
(but as said this is only my guesses I don't know what the standard states).

EDIT2------------
I had to experiment with the BIAB load special settings: ”BB song bar offset” and ”Number of bars in XML to skip” in order for the chords to align correctly in BIAB. It seems to depend on if there is a non complete bar at the beginning of the song. The setting that worked for me was ”BB song bar offset=2”, ”Number of bars in XML to skip=1” and the first bar was only 1 beat (in the original song from SongMaster Pro via MuseScore). The file that I tested had chords spread over various beats 1,2,3,4 and they seemed to align correctly. Only thing it will miss the first bar with only one beat.

EDIT3 ---
It seems to work fine with the default settings as well ie ”BB song bar offset=2”, ”Number of bars in XML to skip=0”. Only thing is that the 3 first beats will get an C chord since my song only begin with one beat in the first bar.

Below attached are screen shots:
1) SongMaster Pro first 20 bars
2) Imported MusicXML from SongMaster Pro into MuseScore after copied over to new empty score
3 New MusicXML exported from MuseScore and imported into BIAB with default settings ie XML bars to skip=0.

Thanks for the update shlind. I would have expected that Music XML 4 would have been backwards compatible, but perhaps there are other features in there that don't quite match up.

Your experimentation and feedback have provided valuable information.

Back to SongMaster Pro for a bit, which program I've just started exploring.

I'm guessing I know the answer to this and it's "yeah, sometimes it just does that".

I've been trying to analyse the Dave Brubeck band's "Take Five" and it seems to fail fairly spectacularly to do what it should. It thinks it's in 4/4 (OK, easily enough fixed) and unsurprisingly I have to correct the timing at the beginning. It looks like it probably gets the chords right. But it fails to get most of the stems anything like right. Drums are OK, bass may be OK, most other instruments are in "The Rest" apart from some occasional witterings and the odd section where it decides the Sax is a guitar.

I can't say I'm surprised, really, as this always seemed a major "wow, that's incredible" and the instruments all have rich spectral content. It does transpose and adjust tempo OK, which was a large part of what I wanted, and I imagine it will work better for other pieces. I haven't seen anything yet that suggests there's a way to better tune it for generating the stems of other sounds.

Hey Gordon, I'm fascinated by what's going on under the hood with stem separation. I suspect that if the program wasn't trained on songs similar to Take Five, then it probably won't do a good job because of the sonic overlap of the various instruments.

One thing you might try is to make it easier to for the program by slowing down the tempo of the file you feed it . . . just a thought.
Another thought is to email the file to the developer and see if they can do any better. At a minimum you'll makke them aware of the kind of music you're trying to decompose.
--Steve

I've been using the program intensively during the last months, and it's by far, the best buy I've made in many years, I'm saving sooooooo much time when transcribing songs / entire repertoires.

The program, however, is not perfect, and still have its limitations. I'm not sure which stem separation algorhtym is used by the program, but in any case, the algorhtym has been trained to meet the demands of most musicians and music aficionados, which means that works REALLY well with mainstream music (essentially modern pop and rock tunes), and not so well with jazz, and other not so popular styles.

It also works pretty well with modern mixes (from 1970s onwards), but not so well with older songs where, for example, the bass may be panned all the way to the right, while the drums are panned all the way to the right.

Regarding chord recognition, the program, again, does a great job with simple harmonic material (pop, rock, blues, soul, funk tunes) but may have some problems with more advanced, jazz based, compositions.

I've also found that the quality of the source has a great impact on the results: just avoid 192khz and below MP3s.

Knowing these limitations is essential to get the most out of the program, and avoid getting frustrated.

Steve ... I'll let AurallySound know, though I'm a bit sceptical how much they can improve things. I'll be interesting to find out.

Chord recognition on jazz is very exuberant. Not surprisingly it picks up on all sorts of stuff that jazz players put in. I knew I'd have to rationalise.

The observation about bit-rate is helpful ... I hadn't considered it, though now I'm informed it's obvious.

What I'm asking of it is quite a lot and I'm far from surprised it didn't make it. It will still be useful for things like transcription, though sometimes harder work. That's life; and it'll likely get better with tech advances.