PG Music Home Forums PG Music Wishlists Band-in-a-Box Wishlist Styles that actually sound something like the style name &/or band/song title linked to it.

I find this subject absolutely fascinating in part because of the complexity it embodies. In my career I’ve needed to deal with, embrace and at times, crush complexity so that the goals of the project could be met. Thousands of people do this daily across most any domain.

It appears that for this issue to truly be resolved, a technical definition for what a genre/style is, compatible with machine coding or machine searching has to be developed. For the Berklee Musicologists and other experts out there do you think developing such a crisp definition is even possible, say in the next 5 years or so? I have no doubts that at some point this will be solved.

I’ve pondered this question for only a few days now and I’m gaining a deeper appreciation for what the programmers at PG Music face when trying to create a robust Style Picker. I wouldn’t mind informally working on this problem with others. But I suspect we will face a wall very quickly when we realize that styles and genres are subsets of music and music embodies human emotion and various cultural contexts. Defining emotion and context algorithmically or even with neural nets or random forests will require much inter-disciplinary brain power. 1300 genres may not seem like a big number until you try to encode your 1st genre . . . like I say, fascinating subject.

The following is a small section taken from an article at ProMusicianHub.com, I think it’s a good read to help get your arms around this topic.

https://promusicianhub.com/what-is-music-genre/

“A music genre is a classification system that classifies music into different styles. It’s the art of incorporating instrumental and vocal tones in a structured manner that gives the music its distinctive character. As a result, all artistic compositions that belong to the same music genre share some similarities in form or style. The word genre is used in other forms of art, including literature, television, cinema, and other artistic creation types. It combines pieces of work that fit under a specific category after analyzing and highlighting the most distinctive elements.

In addition to pure genres, crossover genres are also popular. These combine several elements from different elements to create a new style that appeals to a wider audience. This technique is widely used in popular music. However, it’s quite common for a musical piece to belong to several genres at the same time. Music genres can refer to when the music was composed, its style, the instruments used and their functions, and the music’s geographical origin. In other words, there’s no one clear comprehensive definition that explains what different music genres are. Even within a single system, there is some disagreement about the interpretations of the elements used to categorize music composition.

For example, some researchers state that the Classical period which marks the creation and popularity of Classical music, lasted from 1730 or 1750 to 1830. However, some researchers say that this period started in the 1780s or later in the 1790s. Both interpretations are accepted.”

As a former programming professor, I concur that defining the problem is the most important step. There actually has been a lot more discussion about the style picker among some of us who test the program, and the developers participate.

As I mentioned above, I usually don’t even use the style picker. My compositions cross genres.

For those interested in what makes up the psychological groove factor in a song/drum pattern. I found an interesting research article. It may give some input to what to listen to in the music when trying to match a suitable style with a resemblance in the feel/grove of a song. I found the parameters as drum pattern category, syncopation and event density to be particular interesting. One problem now is how I will be able to hear variances in this in the music I am working on.

Link to the research article “Groove in drum patterns as a function of both rhythmic properties and listeners’ attitudes (the paper can be downloaded as PDF)
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6025871/

All the input data, drum patterns, notation, midi and MP3 are available for download if anyone is interested.
https://www.grooveresearch.ch/index.php?downloads

Thanks for sharing this paper, I read the abstract and introduction. I would never have thought that groove itself would be a serious topic of research. “The empirical study of groove as a research field within music psychology, however, only dates back to the early 2000s.” More power to these researchers in Switzerland.

My vision for this overall subject is to one day be able to mix 2 different styles to produce a new style. We probably all remember when Grunge emerged from Seattle in the 1980s from a fusion of Punk and Heavy Metal. How cool would it be if you heard an awesome new style/genre for the first time . . . that you created on your desktop computer by blending 25% Reggae with 75% Red Dirt Country?

Science fiction you think? You’re right. And at one point in the past both BiaB and AI were science fiction, yet both are readily available today.

Can't you do that right now in BiaB? Take tracks from different styles to create a new style? I don't know if you can save a new style in BiaB as I have never tried that and today life is getting in the way so I can't try it myself. But I do know that you can take different tracks and save them in your DAW; that would give you a new style but you won't be able to save it as a BiaB style.

Mario, I don't think BiaB or any other program can do this at this point in history, that's why it's a vision. I'm not talking about exporting stems, tracks or songs from BiaB and manually mixing them in a DAW.

Rather, this is about creating a brand new style/genre that you've never heard before by mixing/blending 2 existing styles algorithmically aka in software, via digital signal processing and other automated operations.

For example, 20 somethings (plus perhaps a little recreational chemistry) fused Punk and Heavy Metal to produce Grunge. As a test, I'd like software that can repeat this and a few other examples to convince me that it can be done; then I'd ask it to fuse other styles that haven't been fused yet such as Reggae and Red Dirt Country.

Not human-in-the-loop, but software-in-the-loop per the block diagram.

If I'm wrong (and I hope I am), what page of the BiaB manual says that this can be done?

You can kind of mix two styles now. The limitation is you can't play two styles simultaneously. MultiStyle plays back the audio from either one style or another style. What doesn't blend together is the artificial intelligence that is unique and built-in to each style. To me that would be like hearing two people try to talk at the same time.

Of course, when we talk about mixing two different styles to produce a third, we're talking about producing a third style that is pleasing to the musician and non-muscian ear. Something that would produce a following. Otherwise, what's the point in producing noise?

From Wikipedia "Grunge (sometimes referred to as the Seattle sound) is an alternative rock genre and subculture that emerged during the mid-1980s in the American Pacific Northwest state of Washington, particularly in Seattle and nearby towns. Grunge fuses elements of punk rock and heavy metal, but without punk's structure and speed."

It's funny. I don't particularly care for Punk or Heavy Metal but when elements of these two were "mixed" I do like the sound of Grunge. Go figure.

I just found this online AI-service tool for extracting instruments from an audio file.

LALALAI
https://www.lalal.ai/

I have only played around with the free version extracting out various instruments drums, bass etc.
I am impressed, I have tried several ways to extract drums from mp3 before e.g. various EQs,inversion, Drum Remover plug ins but nothing have produced anything useful at all.

This AI tool could be of help for a user when analysing a song arrangement trying to hear the various instruments in an existing arrangement.

It is very simple to use just select which instrument to separate and drop the the mp3 file in the browser. The free version does only give a sample of the song and it is only for listening no file produced. (But it is quite easy to record what is played back on the PC eg with Audacity with the interface set to loopback what is played in the speakers and record it (Windows WASAPI)).

While waiting for improvements of the search concept in the style picker, here are some ideas and suggestions for some improvements in the style picker search mechanism:

1.Style Picker GUI: Make it more efficient to review a selected set of styles.
Today the user have to:
a) select the style
b) klick on the Play button (against my chords or the demo)
c) go to a) above

This is very time consuming, a suggestion is:
I. Allow a selection of styles, by marking up and selecting them one by one or as chunks (e.g. by use of the shift key as in most other software e.g. like marking up text rows in MS Word).
II. Allow the user to step to the next style in the marked up list and play it by a simple klick on Play.
This will allow users to review the styles more efficiently just by making a selection of styles, then only to klick Play to step through the list and play the styles one by one.

2. Real Drums Picker GUI: Make it more efficient to review a set of drum styles.
I. Basically it is the same problem in the Real Drums picker GUA as in the Style picker GUI. It has the same problems i.e. the user have to select one item from the list then klick play, then select another item, then klick play etc, very time consuming (improvement as suggested above).
II. Another improvement would be to be able to a tick box to “collapse” all variants of the same drum patterns. This will provide a more compact view of the overall drum patterns when trying to find the overall right feel in the drums patterns.
III. Another improvement would be to have an option to go back and forth from the current marked up selection in the style picker and having a drum style selected in the marked up drum list. Then the user can continue to review the narrowed down styles in the selected styles list that contains the particular drums. I know it is possible to add a search criterion in the style picker for this but it would be so much easier and more efficient if one could go back and forth between the style picker and drum picker keeping the selections and trying out various items from the selections. I also know that the drum picker lists the styles that have a particular drum style in a tiny window, this is of some help but it is not efficient having to go back to the style picker to check out those styles one by one and in between each review have to look up the drum styles again to check the list of styles unless I write them down and tick them of manually.

3. General point of table search in the GUIs. A general point to narrowing down the search in tables in the style picker and drum picker and in searching tables in general:
I. Provide a “hierarchy column sort”. I.e to be able to sort one column, this is possible today, then mark up a selection of items from that column and continue to make a sort of another column within that selection (keeping the previous column selection) etc. That is how most table searching GUIs work. In BIAB today all previous column sort orders disappears when sorting a new column.


4. Help to find a good match to the Feel/Groove of a particular song:
The current song title search function as today only helps out by setting four of the filter parameters “Cathegory, Time Signature, Feel, Tempo”. It does not rank or sort the list in any order for how well the styles reassembles the Feel/Grove of the song at all. If I make a test and put in these four parameters manually I will get the exact same result i.e. the same list in the same order without any connection to the Feel/Groove of the song. It is a good start to get this help in setting these filter parameters by the software (but on the other hand it is not very difficult to listen to the song and set them manually).

I think the Drum Pattern Matrix is one major component in order to find something that sounds reasonable similar to the song to match the feel/groove of.
And that is more or less what Henry Clarke uses in his work flow as mentioned in one of the posts and links above, but he does it manually by ear and by help of a midi file.
I. It would be very useful if BIAB could provide some level of support for this in the software i.e. “Best Matching Drums” at least for the case where the user can provides a midi file including the drums.
Preferably resulting in a list of Real Drums/ Midi Drums in Ranking order (like EZDrummer %match). Any level of help framing it down would be useful.
Then from that list of best matching drums a simple way back to the filtered list of styles with all the styles that have the selected drum pattern marked up (as mentioned in 2. above).

II. To match the drums in an mp3 song file is probably not an easy task. But maybe it is possible to provide some simple function for the user to “tap in the basic feel” and the BIAB then helps out sorting out the best and most likely matches from the drum list in the same way as for the midi case above.