Skip to main content ↑ ↓

Notice

Please note that most of the software linked on this forum is likely to be safe to use. If you are unsure, feel free to ask in the relevant topics, or send a private message to an administrator or moderator. To help curb the problems of false positives, or in the event that you do find actual malware, you can contribute through the article linked here.
Topic: Musical texture digital detector (Read 7375 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

Musical texture digital detector

Does anybody know such a software as in the thread title?

What I'm looking for is basically a software that tells me what kind of instruments are played in a music track without the need to play it and listen to it, just by analyzing the waveform.
Does it exist?
I'm late

Musical texture digital detector

Reply #1 –
I don't know of any such software and I don't believe that's possible.

It might be possible for software to identify a single instrument, but I wouldn't expect good accuracy...  I'd expect lots of errors.

With more than one instrument playing at the same time things get very complicated very quickly.

And of course if there are electric guitars with effects that change the nature of the sound or "artificial" synthesizer sounds there's no way for software to understand the unlimited number of possibilities.

Musical texture digital detector

Reply #2 –
It is just like the old "how can I convert mp3 to MIDI?" and "How to remove vocal or some specific instruments from a song?" question. No software can do such things automatically with acceptable quality, at least for most acoustic instruments.

Musical texture digital detector

Reply #3 –
Quote from: bennetng on
No software can do such things automatically with acceptable quality, at least for most acoustic instruments.


Unacceptable quality would do, at least to start off with. I need something rough to work on. I'm pretty sure people who analyze big data from online streaming services - I'm thinking of the guys at Echonest especially - have tools that translate musical texture in measurable values.
I don't expect to find a software that tells me exactly what the music ensemble is, but at least to have a parameter associated with the overall timbre, to experiment with.
I'm late

Musical texture digital detector

Reply #4 –
I just read Echonest's website. What I understand is they are not solely depend on the audio signal to identify music. They make use of listener's personal profile and various kinds of metadata as well. This approach can be useful if a particular piece of music/song is being listened among many people. Plus there are watermark technologies to facilitate music identification. But let's say if I composed a piece of original music that no has heard before, can it describe my music accurately? Maybe it can still identify the genre if it is a popular one, but for individual instruments I really doubt, as suggested by DVDdoug already.

[EDIT]
Did you know about SoundHound? It can quickly identify a piece of popular Final Fantasy music from my phone's microphone:
https://youtu.be/jLbULEiuFgo

But it cannot recognize a transcribed one
http://d.hatena.ne.jp/video/niconico/sm9917356

The transcribed one is done by me. Yes it is far from perfect but I've uploaded it to youtube before, the result is that I got a lot of comments, PMs and emails and they wish I can upload the MIDI data. However, some of my previously uploaded MIDI data were stolen by some shameless people and they reused my data to make rearrangements and remixes without credits so I decided not to upload any MIDI data anymore. I got frustrated that people keep asking and finally I deleted my youtube account. Therefore I am still pretty confident that my transcriptions are at least easily recognizable by human.

Musical texture digital detector

Reply #5 –
Quote from: davideleo on
Quote from: bennetng on
No software can do such things automatically with acceptable quality, at least for most acoustic instruments.


Unacceptable quality would do, at least to start off with. I need something rough to work on. I'm pretty sure people who analyze big data from online streaming services - I'm thinking of the guys at Echonest especially - have tools that translate musical texture in measurable values.


Wikipedia has links to the dissertations their technology was based on.  I suppose thats a good place to start. 

Musical texture digital detector

Reply #6 –
Quote from: bennetng on
Did you know about SoundHound? It can quickly identify a piece of popular Final Fantasy music from my phone's microphone:
https://youtu.be/jLbULEiuFgo

But it cannot recognize a transcribed one
http://d.hatena.ne.jp/video/niconico/sm9917356

The transcribed one is done by me. Yes it is far from perfect but I've uploaded it to youtube before, the result is that I got a lot of comments, PMs and emails and they wish I can upload the MIDI data. However, some of my previously uploaded MIDI data were stolen by some shameless people and they reused my data to make rearrangements and remixes without credits so I decided not to upload any MIDI data anymore. I got frustrated that people keep asking and finally I deleted my youtube account. Therefore I am still pretty confident that my transcriptions are at least easily recognizable by human.



I heard about it, but I never used it.
Your transcription actually sounds pretty similar to the original. So, you're telling me SoundHound can actually tell the difference between a live orchestra and a synthetic one. Interesting! Thanks for the hint.
I'm late

Musical texture digital detector

Reply #7 –
Quote from: saratoga on
Wikipedia has links to the dissertations their technology was based on.  I suppose thats a good place to start.


I'll have a look, thanks.
I'm late

Musical texture digital detector

Reply #8 –
Quote from: davideleo on
I heard about it, but I never used it.
Your transcription actually sounds pretty similar to the original. So, you're telling me SoundHound can actually tell the difference between a live orchestra and a synthetic one. Interesting! Thanks for the hint.


By saying my transcription is "far from perfect" I mean some of my chords, harmonies, harp glissando scale etc are different from the original. It is related to my musical ability, not about the quality or realism of expression. In fact Soundhound allows users to hum a melody to identify a song, which means their analysis is based on the structure (tempo, pitch, harmony) of a song rather than "musical texture" as you described.

Not a surprising result that machine transcriptions failed on multitimbral or highly polyphonic music.

[EDIT] I've uploaded a monophonic midi file for those who are interested. For consistency please use Windows' default MSGS wavetable synth to play the file if possible.
http://www.hydrogenaud.io/forums/index.php...ost&id=8288

Soundhound can identify the above file as this one regardless of timbre:
https://youtu.be/ka9qs4nr92M

Musical texture digital detector

Reply #9 –
deleted

Musical texture digital detector

Reply #10 –
I've been thinking of examples, and if you wanted this software to detect a saxophone, it would have to pick up the coma-inducing smoothness of Kenny G and identify it as the same instrument as the guy with that horribly nasty blaring tone on Pink Floyd's 'Delicate Sound of Thunder'. And his regrettable mullet.

The day a piece of software can do that is the day before SkyNet becomes self-aware...

Musical texture digital detector

Reply #11 –
Quote from: JabbaThePrawn on
I've been thinking of examples, and if you wanted this software to detect a saxophone, it would have to pick up the coma-inducing smoothness of Kenny G and identify it as the same instrument as the guy with that horribly nasty blaring tone on Pink Floyd's 'Delicate Sound of Thunder'. And his regrettable mullet.


Well, no. I'm not actually interested in detecting the single instruments involved, but rather in classifying the physiological impact of different ensembles. But "coma-inducing" Kenny G is priceless 
I'm late

Musical texture digital detector

Reply #12 –
This article states that researches have been using "signal processing to analyse the musical properties of songs" such as chord changes. Any idea of what applications can do this?
I'm late

Musical texture digital detector

Reply #15 –
Actually it's theoretically possible to do using the same technology that can now identify birdsong or human songs. No one's done it though as well... there's not a lot of need.

Musical texture digital detector

Reply #16 –
I still remember around 2002, I debated with someone about this "automatic polyphonic multitimbral music transcription" issue and he used speech and handwriting recognitions as examples. I told him "let's say at a bus stop, when everyone is talking, can a program recognize the speeches of EACH person?". He was speechless and just told me MPEG-7 can do perfect music transcription in less than 7 years. Now it is 2015 already and nothing revolutionary happened. At that time the best transcription program I could find is SONIC, but it is only limited to piano-like sounds.
http://lgm.fri.uni-lj.si/matic/SONIC/

Melodyne's DNA looked nice in their demo video but it is not really better than SONIC in terms of pitch detection quality. Also note that DNA is not supposed to use with multibimbral materials.
http://helpcenter.celemony.com/editor2/en/...r_30.html?q=dna

From my experience, such programs work best with sounds contain distinctive attack and "connectable" sustain and release, as well as easily recognizable harmonic patterns, that means you can clearly see some continuous horizontal lines, maybe with some moderate vibratos and tremelos in their spectrum graphs.

Maybe we can wait another 7 or 14 years, but if it really happens it almost means machines can think, learn and behave like human, I am really afraid of it.