Add support for Elektron Syntakt sample filesystem

[dagargo]

With Syntakt 1.40, a new machine for samples has arrived and I’d like to add support for it to Elektroid.

While I have no idea of the actual MIDI SysEx implementation, some technical details are available and some of its limitations might imply some details of the underlying design.

For instance, it seems that the sample filesystem is similar to the projects and sounds ones as it is a slot filesystem where items are addressed by their ID. Another detail confirming this would be that changing the order of the samples would affect any project using that slot. Additionally, we know there are 64 samples whose maximum length is about 5 s. Besides all this, there is no project or RAM management and samples are configured globally.

All in all, it might be that samples can be addressed as all data filesystem, like projects and sounds, and its content is directly load into RAM at startup.

Here is something for anyone interested to try. Just list the data filesystem and cross your fingers.

$ elektroid-cli ld
0: id: SYSTEM_ID; name: computer
1: id: hw:3,0,0; name: hw:3,0,0: Elektron Digitakt MIDI 1

$ elektroid-cli elektron-data-ls 1:/
D 0000 0 0         0B  -1 projects
D 0000 0 0         0B  -1 soundbanks

This is just wishful thinking and it might be totally wrong but we’ll see.

[kassoulet]

Not so easy I guess. I tried to activate "sample" and "storage" in the Syntakt entry, but it's not working.

Good news is, I can actually see the samples in data mode:

$ elektroid-cli elektron:data:ls  2:/
D 0000 0 0         0B  -1 projects                        
D 0000 0 0         0B  -1 soundbanks                      
D 0000 0 0         0B  -1 samples     

$ elektroid-cli elektron:data:ls  2:/samples
F 007e 1 0   75.79KiB 001 BD BAUTA                        
F 007e 1 0   29.03KiB 002 BD DUNK                         
F 007e 1 0   29.73KiB 003 BD KLASSIK                      
F 007e 1 0   79.17KiB 004 BD KRUSH                        
F 007e 1 0   85.45KiB 005 BD LEE          

[dagargo]

Thanks for trying this out, @kassoulet!

This is exactly what I was expecting and it's really great because we already have most of the code needed to implement this.

Do you want to try to implement it with some help? Or should I proceed with the creation of the filesystem?

I'll need help anyway with the upload and download code and testing.

[kassoulet]

I'll try this weekend to implement this. It seems easy, samples on syntakt are handled like the projects. I thought using the same setting as other samplers would work, but nope, that a pity.

Also, the Elektron Tranfer app display a message to explain that samples are shared over projects and that changing or removing them will affect existing projects. Not sure about displaying in the app or in the doc :)

[dagargo]

I've updated the description with the same comment I posted on Elektronauts yesterday.

[dagargo]

I'll try this weekend to implement this. It seems easy, samples on syntakt are handled like the projects. I thought using the same setting as other samplers would work, but nope, that a pity.

I've done some tests very early today and I think it's gonna be easy. The only thing I'm unsure is the metadata that might be written together with the sample data, such as duration, loop points...

I might end up pushing it as it might be faster. But some work will be required for sure.

Also, the Elektron Tranfer app display a message to explain that samples are shared over projects and that changing or removing them will affect existing projects. Not sure about displaying in the app or in the doc :)

I'd rather forget about that message completely.

[dagargo]

I've just pushed some generic code that I think it might work except for uploading and downloading, which will require some adjustments for sure. It's in the branch feat-add-syntakt-samples.

I'd recommend you to try to download some samples and see if you can figure out the whole payload. Then you'll need to do the same for the uploading part.

[kassoulet]

I got the sample to be downloaded. BUT there is junk before and after it. Not the same as the brothers, I'll have to reverse it, at least to get the real sample size.

[dagargo]

I was expecting something like that.

Perhaps I haven't understood you but the junk before might follow the same structure the others samples have. Take a look at the elektron_sample_header struct in the elektron.c file.

If that's right, then you'll have the length and be able to truncate the junk at the end.

[kassoulet]

Well no it's not. Even worse, with bigger samples, the data clearly comes with several packets, missed by the actual code, ending with un-aligned reads of two-bytes data, it's a total mess. I'll go for a walk and retry later :)

Here is the start of one dump of received data. Audio starts at 0xb0:

│00000000│ ac 11 d3 03 02 00 08 00 ┊ 0d 30 30 38 32 00 00 00 │וו•0•0┊_0082000│
│00000010│ 08 ff ff ff ff 00 00 00 ┊ 02 00 00 74 22 01 0c 00 │•××××000┊•00t"•_0│
│00000020│ 00 73 93 51 53 41 4d 50 ┊ 00 01 00 91 39 f1 42 44 │0s×QSAMP┊0•0×9×BD│
│00000030│ 20 44 55 4e 4b 0e 00 0f ┊ 02 00 17 f3 ff ff ff ff │ DUNK•0•┊•0•×××××│
│00000040│ ff ff ff ff ff ff ff ff ┊ ff ff ff ff ff ff ff ff │××××××××┊××××××××│
│*       │                         ┊                         │        ┊        │
│000000a0│ ff ff ff ff ff ff ff ff ┊ ff ff ff ff e1 ff fe 00 │××××××××┊×××××××0│
│000000b0│ 39 02 f9 0f 85 2b 67 4a ┊ b3 57 e2 53 45 4e c6 45 │9•×•×+gJ┊×W×SEN×E│
│000000c0│ 75 2f bc 1c 03 04 f9 eb ┊ 9e d8 35 c6 a5 b8 e5 b0 │u/ו••××┊××5×××××│
│000000d0│ f3 ab 21 a9 ca ad 4a b0 ┊ 84 b9 bc c3 e4 d0 d1 e0 │××!×××J×┊××××××××│
│000000e0│ 04 ee 56 fe 66 0d 9b 1b ┊ 7d 28 ec 33 3c 3b ba 43 │•×V×f_ו┊}(×3<;×C│
│000000f0│ 0a 48 09 4c cd 4d cb 4d ┊ cc 4b 5a 47 20 41 7b 39 │_H_L×M×M┊×KZG A{9│
│00000100│ c1 30 ed 27 8f 1d 07 12 ┊ fc 07 c2 fc e8 f2 23 e7 │×0×'ו••┊ו××××#×│

Nothing looking that the size of transfered or final sample. The end of the sample is silent and cropped. Audio is 48000khz.

[dagargo]

Doesn't look great.

Could you post somewhere the downloaded payload (data file) and the uploaded sample with Transfer here so I could take a look at both?

Edit: If it's of any help, Elektron machines are big endian.

[dagargo]

Actually, you have the size of the payload in the output ls command. If you remove the 0xb0 sized header, does it match the size of the sample data?

$ elektroid-cli elektron:data:ls  2:/samples
F 007e 1 0   75.79KiB 001 BD BAUTA
[...]

[dagargo]

I've been examining the data and it's very difficult to know what's the underlying structure.

When you run your tests, you didn't use Elektroid to convert your samples to 48 kHz before copying them with Transfer, did you? By reading your posts, it looks like you didn't but I might be wrong.

Would it be possible for you to use Elektroid to convert your samples to 48 kHz and then transfer them with Transfer and finally download the raw data with Elektroid? It's difficult to know what's going on but, on top of this, a different resampling algorithm is used and the samples and amount of frames are different. Or so I think.

To convert the samples, use the local filesystem and the mono 48 kHz 16 bits filesystem.

If possible, send me the converted samples by Elektroid and the raw data files.

Thanks in advance.

---

Position 0x19 seems to be a u32 storing something like the full size or the frames * 2.

$ hexdump -C -v BD_DUNK_STK.raw | head -n 20
00000000  ac 11 d3 03 02 00 08 00  0d 30 30 38 32 00 00 00  |.........0082...|
00000010  08 ff ff ff ff 00 00 00  02 00 00 74 22 01 0c 00  |...........t"...|
00000020  00 73 93 51 53 41 4d 50  00 01 00 91 39 f1 42 44  |.s.QSAMP....9.BD|
00000030  20 44 55 4e 4b 0e 00 0f  02 00 17 f3 ff ff ff ff  | DUNK...........|
[...]

[kassoulet]

The samples are 44100 Hz WAVs on disk. I sent them using Transfer. And I downloaded them using Elektroid. I noticed they were not the same size, and loaded with Audacity in 48000 Hz they were alligning perfectly. So I presume Syntakt also store them in 48k. But I'm not sure.

The settings used to save the downloaded samples:

static const struct fs_operations FS_DATA_SAMPLES_OPERATIONS = {
  .id = FS_DATA_SAMPLES,
[...]
  .name = "data-sample",
  .gui_name = "Samples",
[...]
  .download = elektron_download_data_sample,
[...]
  .save = file_save,

I'm still not convinced. If this is real, Elektron made a 44.1 kHz sample pack, resampled to 48 kHz by Transfer, this is insane :)

ps:
From Syntakt manual: "the native sample file format is 16-bit, 48 kHz, mono .wav files"
So it's confirmed, Elektron didn't bother to read their own manual :)