Vintage Timestretching PLEASE Akai

Be amazing if they could implement the famous Akai s950/s1000 timestretching algorithms.
I wonder if they could still get the code from somewhere, do you think they have the code still?

In the meantime there is the Akaizer software which is very nice but as good imo
https://the-akaizer-project.blogspot.com/

I'd have to look it up for the s1000 because it is different, but the s950 does not use a time stretch algorithm. Instead what it does is far more clever, but it turns out to be harder to replicate in software for a number of reasons.

When sampling on the s950 you have a variable bandwidth rate from 3000 to 19200, which equates to a sample rate of whatever bandwidth you select multiplied by 2.5 to get the sample frequency. (i.e. sample rate = bandwidth * 2.5). This nets you 19200-3000, or 16200 different potential sample rates.

When playing back a sample on the s950, instead of using an algorithm to change the sound based on pitch, it just alters the speed of the chips that play back the sound. There may also be some anti aliasing features in the chips, but for that I am not sure. The s950 has 8 voices because it has 8 discrete chips that are handling all of the audio playback, and can change the speed each chip plays back audio.

Sure you can simulate a lot of this with DSPs and such, but I think that also ends up missing the point. A lot of the reason that the s950 has a distinctive sound is because of the limits of the hardware, the number of voices, amount of memory, the workflow of the S950, and time it took to actually construct a program... which lead to sound designers making decisions because of the limits of the S950, and what they were trying to accomplish. Working with the S950, you are constantly trying to figure out how to squeeze the most out of the limited memory you have by juggling bandwidth settings, but still maintaining a sound that is recognizable. I am not sure modern producers are looking for the kind of constraints that an S950 puts on them... and to make a plugin that is hard to use, has major constraints on it... well it sounds like a limited market.

I suspect this is why we see so many plugins that sound similar, but don't really put the pressure on the musician. You can get plenty of 12-bit effects, or effects that drop sample rate... but the actual workflow of the S950 is part of what gives it a character... and I don't think most people would put up with it.

All that said, I do get that there's a market for it, but how viable it is I am just not sure.

NearTao wrote:I'd have to look it up for the s1000 because it is different, but the s950 does not use a time stretch algorithm. Instead what it does is far more clever, but it turns out to be harder to replicate in software for a number of reasons.

When sampling on the s950 you have a variable bandwidth rate from 3000 to 19200, which equates to a sample rate of whatever bandwidth you select multiplied by 2.5 to get the sample frequency. (i.e. sample rate = bandwidth * 2.5). This nets you 19200-3000, or 16200 different potential sample rates.

When playing back a sample on the s950, instead of using an algorithm to change the sound based on pitch, it just alters the speed of the chips that play back the sound. There may also be some anti aliasing features in the chips, but for that I am not sure. The s950 has 8 voices because it has 8 discrete chips that are handling all of the audio playback, and can change the speed each chip plays back audio.

Sure you can simulate a lot of this with DSPs and such, but I think that also ends up missing the point. A lot of the reason that the s950 has a distinctive sound is because of the limits of the hardware, the number of voices, amount of memory, the workflow of the S950, and time it took to actually construct a program... which lead to sound designers making decisions because of the limits of the S950, and what they were trying to accomplish. Working with the S950, you are constantly trying to figure out how to squeeze the most out of the limited memory you have by juggling bandwidth settings, but still maintaining a sound that is recognizable. I am not sure modern producers are looking for the kind of constraints that an S950 puts on them... and to make a plugin that is hard to use, has major constraints on it... well it sounds like a limited market.

I suspect this is why we see so many plugins that sound similar, but don't really put the pressure on the musician. You can get plenty of 12-bit effects, or effects that drop sample rate... but the actual workflow of the S950 is part of what gives it a character... and I don't think most people would put up with it.

All that said, I do get that there's a market for it, but how viable it is I am just not sure.

Yup. You really got the most distinctive sound out of the S950 when time stretching a pitched down sample recorded at 45rpm +8% in a lower bandwidth.

I have multiple tools in my arsenal to bitcrush and lo-fi, but it always takes other tools in the mix as well to get a desired result.

The S1000 has been done pretty well and ended up as freeware called Akaizer (sure you know).

I have been wondering why Akai doesn’t simply tap into that source, either licensing the code or just using it if the code is open... especially since they are also oriented towards dance music. Imagine UK electronic music without s1000 timestretch...

Oh and since we’re at it, they should also license Decimort2. But I guess they try to invest as little money as possible into the software and therefore just stick with what they have which means more Air plugins...

To me there is nothing special about the 950 timestretch but thats probably because I first came in touch with timestretch by an article in a keyboard magazin in ...the early/mid 90s and the way it worked was AFAIK the same on all samplers that had this feature.

The sampler cut the long sample into very small pieces (the size of those pieces are adjustable. The 950 call it 'D-Time', CoolEdit/Adobe Audition calls it 'Window-Size')
Lets number those pieces so it's more visual
1,2,3,4,5,6,7,8,9
After that the pieces are played back looped. Depending on how much timestretch is set (and the size of the pieces) the number of the repeated pieces varies
So the playback of the pieces looks like this

1,1,2,2,3,3,4,4,5,5,6,6,7,7,8,8,9,9
or
1,1,1,1,2,2,2,2,3,3,3,3,4,4,4,4,5,5,5,5,6,6,6,6,7,7,7,7,8,8,8,8,9,9,9,9

We all know the issue of clicks in loops caused by 'value-jumps' between loopstart and endpoint. To avoid this the 'microloop-pieces' have 'microfades' and/or small overlapping.

peace

peterpiper wrote:To me there is nothing special about the 950 timestretch but thats probably because I first came in touch with timestretch by an article in a keyboard magazin in ...the early/mid 90s and the way it worked was AFAIK the same on all samplers that had this feature.

peace

You don't know what your talking about, there weren't any other samplers with time-stretch when the S950 came out!
You also have been oblivious to entire genres of music that have come and gone since then, you need some sort of special ignorance award for this comment.

simon harris wrote:You don't know what your talking about, there weren't any other samplers with time-stretch when the S950 came out!
You also have been oblivious to entire genres of music that have come and gone since then, you need some sort of special ignorance award for this comment.

Just curious but - how did his comment suggest he is oblivious to entire genres of music that have come and gone since then? He merely stated how he first learned about the feature in the 90's, and how he perceived time-stretching algos work. Also, by the time the S950 finished its run in 1993, there were other samplers that did time-stretching (Akai's own S1000 as well).

Here's a laborious workaround from Tubedigga that produces good sounding results, but it's a bit fiddly and likely requires adjustment for each source sample:

hyperbolist wrote:Here's a laborious workaround from Tubedigga that produces good sounding results, but it's a bit fiddly and likely requires adjustment for each source sample:

I was just thinking about the s series a few days ago and missing its time stretch. Thats a cool technique that didnt cross my mind im gonna try that out

90's: That timestretch algorithm is too dirty!

2021: That timestretch algorithm is too clean!

ace_of_dub wrote:90's: That timestretch algorithm is too dirty!

2021: That timestretch algorithm is too clean!

nailed it!!

ace_of_dub wrote:90's: That timestretch algorithm is too dirty!

2021: That timestretch algorithm is too clean!

Same thing happened with the MPC2000XL. When it was released everyone complained that it had a sterile sound, now it's apparently got a classic vintage vibe. I've even seen the MPC1000 described in similar terms.

What was once core functionality becomes an effect in somebody's utility belt.

Bit rate is the samples-per-second. It stands to reason to efficiently use time stretching that one would use the highest capture rate they could to have extra samples so that during the stretch there isnt the need to "fill in the blanks" with duplicates. If you double your sample rate you can effectively reduce your tempo by 50% and it still not click. anything less than 50% would thus require tripple or quad value bitrate of your target.

I seriously question the sanity of this post.. Eff what your familiar with! Learn about digital audio! If you bounce your samples to 44100 at 50% tempo it's playing at 22050 (whether done virtually or by clock chip). I shouldn't tell ppl this crap so their music comes off trash and i got less competition... MF's be living lo-fi tryina act like its purposeful.

If you stretch and render you are effectively placing a lower bit rate sample into a higher bitrate stream. Polishing a turd wasting bytes and time. Like taking a polaroid of a painting.

If anyone here doesnt know something. it's simon.

Here's one for ya simon from del: https://m.youtube.com/watch?v=XzE7S3na3J0

simon harris wrote:

peterpiper wrote:To me there is nothing special about the 950 timestretch but thats probably because I first came in touch with timestretch by an article in a keyboard magazin in ...the early/mid 90s and the way it worked was AFAIK the same on all samplers that had this feature.

peace

You don't know what your talking about, there weren't any other samplers with time-stretch when the S950 came out!
You also have been oblivious to entire genres of music that have come and gone since then, you need some sort of special ignorance award for this comment.

Much love to you. You seem to really need it. Sorry for trolling you by talking all this bs about timestreching. And many thanks for the award.