http://www.comp.dit.ie/bduggan/music/matt2/index.php Requires Java; you can play into a mic or use a recording, I found out about it just now and tested the one recording, but with the "silence threshold" slider all the way to the left it correctly guessed it was the Old Gray Goose. This was Seamus Thompson playing a fiddle sans accompaniment (aside from scratches on the vinyl...) so don't know if it would have trouble with strummed backing, drums, drones.
"A second website, tunepal.org, has been launched by Dr Bryan Duggan. This is a query-by-playing search engine for traditional dance tunes, with a database of more than 11,500 melodies. If you’ve ever wondered about the name of a tune you have in your head, you can simply play a phrase from the tune into the search engine, and it will search for a match within its database. (Duggan warns that the system works best with “legato”-style instruments such as flute, tin whistle, concertina, accordion, pipes and fiddle).
Feedback from musicians who’ve tried it has been extremely positive so far, with one noting in the traditional music chatroom, thesession.org, that he “tried several more well-known tunes, with good results. It’s very forgiving of minor mistakes, but gets confused if you swing the rhythm too much”. Another remarked that tunepal.org worked well with both obscure and well-known tunes.
All in all, maybe a small step for a musician but a giant leap for traditional music?"
I've got very basic computer skills but I found this software very easy to use. I've played a reel I learnt many years ago off Joe Fitzgerald and did'nt have a name for it but played it into the software once through the A part on my guitar at a moderate tempo through an AKG Cardioid Condenser Mic and it came up with two names "The Burren" or "Michael Cramer's". Many thanks Dr.Bryan, I'll be passing this on to my students.
You know, as a programmer and an academic it is very heartening to find that people use my program. Ive been kind of overwhelmed with all the nice emails I've received since launching tunepal.org. Also it wouldn't be possible at all if it wasn't for the amazing users of this very website contributing tunes. The future for the archival of the canon of traditional music is now in us all contributing, rather than a few dedicated individuals such as Bunting, Goodman, O'Neill and Breathneach.
BTW I am amazed that it worked with the guitar. The transcription algorithm is not designed for stringed instruments!
If tunepal.org is working properly on your computer it will usually get close to 100% accuracy in getting the correct tune. Its hard to make the program work with all the different configurations of microphones out there, hence I have made this troubleshooting guide:
Some observations: The program can handle recordings of solo instruments, of course; like many who post here it doesn't seem to like bodhrans...a murky recording of pipes with guitar threw it off as well; some of the notes came out as "AAAAAAAAAAAA" and the like. It had no idea what Leo Rowsome was playing in one example: My Darling Asleep, semi-murky, constant accompaniment on the regulators of the pipes; but had no trouble hearing past Paddy O'Brien the elder's bass buttons on his box. So murk seems to faze it, albeit some recordings of Tara Diamond playing solo flute worked despite having more murk in the sound (same occasion as the above-mentioned piper).
All of this is immaterial to the program's intended use, but I'm trying to avoid any chore involved with learning these tunes!
It doesn't seem to work with fundamental set to G; does the silence threshold setting affect the transcription process as well? Does the pitch of the instrument being used matter?
Windows users (and others?) will need to make sure the recording level is set on "Stereo Mix"; double click on the volume control in the taskbar, Options>Properties>Adjust playback for recording. You'll need to switch this to record your own live music of course; if you want to make a recording and then play it back it must be on "Stereo Mix".
I'm a bit baffled by how the search engine recognizes tunes out of what sounds to me like Mike "Tubular Bells" Oldfield attempting to write something wind chimes-ish; how does it take the following:
The onset detection algoritrhm (the bit that detects when a new note starts) assumes that the instruments are playing in unison, so it doesnt really work well with accompanyment. Interestingly, pipes with drones present no problem, because the drones are constant. It only detects changes, so notes played on the chanter and regulators get detected.
The pitch detection part tries to find the most prominent harmonic part of the signal.
The matching part uses the edit distance algorithm which counts the number of edits needed to make one sequence of notes into another, so when you see the distance score, thats the number of edits required. It calculates this score for every tune in the database and then returns the top 10 lowest scores. This is the same algorithm used by most successful MIR systems including the reverends excellent abctunesearch.com. I do a lot of preprocessing of the tunes befiore matching though, like making it case insensitive (so that octaves dont matter) removing ornamentation and expanding all the parts.
Transcriber method 2 uses an STFT to detect note onsets. I just put it there as an experiment. Im not sure how effective it is. I should really do an experiment to see if its any use. I suspected it might work better for stringed instruments.
If you want to read all the details you can read my thesis:
I have another algorithm for segmenting sets of tunes called TANSEY by the way
The G fundamental is for a G tin whistle. I will double check it works tomorrow.
The silence threshold is actually very important. If its set too low, it will pick up the sounds of your computer and try and transcribe notes from them. This actually screws up the transcription completely. If its set too high, it wont hear what you play:
tunepal.org - Automated transcription
tunepal.org - Automated transcription
http://www.comp.dit.ie/bduggan/music/matt2/index.php Requires Java; you can play into a mic or use a recording, I found out about it just now and tested the one recording, but with the "silence threshold" slider all the way to the left it correctly guessed it was the Old Gray Goose. This was Seamus Thompson playing a fiddle sans accompaniment (aside from scratches on the vinyl...) so don't know if it would have trouble with strummed backing, drums, drones.
Article about tunepal: http://www.irishtimes.com/newspaper/weekend/2009/0801/1224251814244.html
"A second website, tunepal.org, has been launched by Dr Bryan Duggan. This is a query-by-playing search engine for traditional dance tunes, with a database of more than 11,500 melodies. If you’ve ever wondered about the name of a tune you have in your head, you can simply play a phrase from the tune into the search engine, and it will search for a match within its database. (Duggan warns that the system works best with “legato”-style instruments such as flute, tin whistle, concertina, accordion, pipes and fiddle).
Feedback from musicians who’ve tried it has been extremely positive so far, with one noting in the traditional music chatroom, thesession.org, that he “tried several more well-known tunes, with good results. It’s very forgiving of minor mistakes, but gets confused if you swing the rhythm too much”. Another remarked that tunepal.org worked well with both obscure and well-known tunes.
All in all, maybe a small step for a musician but a giant leap for traditional music?"
# Posted on August 10th 2009 by Kevin Rietmann
Re: tunepal.org - Automated transcription
This member created that, he's been on here talking about it before:
http://www.thesession.org/members/display/23297
# Posted on August 10th 2009 by tnoumarap
Re: tunepal.org - Automated transcription
Ah, missed that when it showed up last month. Oh well, still a very useful tool.
# Posted on August 10th 2009 by Kevin Rietmann
Re: tunepal.org - Automated transcription
Well, gosh. I am so flattered! The Irish Times not only finds me quotable--they even called me a musician!
# Posted on August 10th 2009 by tuckered out
Re: tunepal.org - Automated transcription
I've got very basic computer skills but I found this software very easy to use. I've played a reel I learnt many years ago off Joe Fitzgerald and did'nt have a name for it but played it into the software once through the A part on my guitar at a moderate tempo through an AKG Cardioid Condenser Mic and it came up with two names "The Burren" or "Michael Cramer's". Many thanks Dr.Bryan, I'll be passing this on to my students.
# Posted on August 11th 2009 by Tony O'Rourke
Re: tunepal.org - Automated transcription
You know, as a programmer and an academic it is very heartening to find that people use my program. Ive been kind of overwhelmed with all the nice emails I've received since launching tunepal.org. Also it wouldn't be possible at all if it wasn't for the amazing users of this very website contributing tunes. The future for the archival of the canon of traditional music is now in us all contributing, rather than a few dedicated individuals such as Bunting, Goodman, O'Neill and Breathneach.
BTW I am amazed that it worked with the guitar. The transcription algorithm is not designed for stringed instruments!
If tunepal.org is working properly on your computer it will usually get close to 100% accuracy in getting the correct tune. Its hard to make the program work with all the different configurations of microphones out there, hence I have made this troubleshooting guide:
http://www.comp.dit.ie/bduggan/music/matt2/troubleshooting.php
Bryan
# Posted on August 11th 2009 by skooter500
Re: tunepal.org - Automated transcription
Some observations: The program can handle recordings of solo instruments, of course; like many who post here it doesn't seem to like bodhrans...a murky recording of pipes with guitar threw it off as well; some of the notes came out as "AAAAAAAAAAAA" and the like. It had no idea what Leo Rowsome was playing in one example: My Darling Asleep, semi-murky, constant accompaniment on the regulators of the pipes; but had no trouble hearing past Paddy O'Brien the elder's bass buttons on his box. So murk seems to faze it, albeit some recordings of Tara Diamond playing solo flute worked despite having more murk in the sound (same occasion as the above-mentioned piper).
All of this is immaterial to the program's intended use, but I'm trying to avoid any chore involved with learning these tunes!
It doesn't seem to work with fundamental set to G; does the silence threshold setting affect the transcription process as well? Does the pitch of the instrument being used matter?
Windows users (and others?) will need to make sure the recording level is set on "Stereo Mix"; double click on the volume control in the taskbar, Options>Properties>Adjust playback for recording. You'll need to switch this to record your own live music of course; if you want to make a recording and then play it back it must be on "Stereo Mix".
I'm a bit baffled by how the search engine recognizes tunes out of what sounds to me like Mike "Tubular Bells" Oldfield attempting to write something wind chimes-ish; how does it take the following:
E,FFGA,B,DA,E,FA,DA,DDddG,cDGGFggeCG,
EdG,A,E,DB,CGeeCDE,EDDCEDDDzG,GAB,
DFA,dCDeGGG,GE,D
and recognize this hornpipe: http://www.thesession.org/tunes/display/8670
What's the difference between transcriber methods, too?
Excellent work Bryan, hope you don't mind all these questions.
# Posted on August 11th 2009 by Kevin Rietmann
Re: tunepal.org - Automated transcription
The onset detection algoritrhm (the bit that detects when a new note starts) assumes that the instruments are playing in unison, so it doesnt really work well with accompanyment. Interestingly, pipes with drones present no problem, because the drones are constant. It only detects changes, so notes played on the chanter and regulators get detected.
The pitch detection part tries to find the most prominent harmonic part of the signal.
The matching part uses the edit distance algorithm which counts the number of edits needed to make one sequence of notes into another, so when you see the distance score, thats the number of edits required. It calculates this score for every tune in the database and then returns the top 10 lowest scores. This is the same algorithm used by most successful MIR systems including the reverends excellent abctunesearch.com. I do a lot of preprocessing of the tunes befiore matching though, like making it case insensitive (so that octaves dont matter) removing ornamentation and expanding all the parts.
Transcriber method 2 uses an STFT to detect note onsets. I just put it there as an experiment. Im not sure how effective it is. I should really do an experiment to see if its any use. I suspected it might work better for stringed instruments.
If you want to read all the details you can read my thesis:
http://www.comp.dit.ie/bduggan/downloads/PhD.pdf
And various papers:
http://www.comp.dit.ie/bduggan/music/index.php
I have another algorithm for segmenting sets of tunes called TANSEY by the way
The G fundamental is for a G tin whistle. I will double check it works tomorrow.
The silence threshold is actually very important. If its set too low, it will pick up the sounds of your computer and try and transcribe notes from them. This actually screws up the transcription completely. If its set too high, it wont hear what you play:
http://www.comp.dit.ie/bduggan/music/matt2/troubleshooting.php
Bryan
# Posted on August 11th 2009 by skooter500
Re: tunepal.org - Automated transcription
Oh BTW a youtube video of MATT2, the predecessor of tunepal.org:
http://www.youtube.com/watch?v=hc0fxPO3u6s
# Posted on August 11th 2009 by skooter500