Darkglass Anagram - NAM profile player & multieffects unit

[Dom Mcsweeney]

Worth a watch:

 

[2dor]

Worth a watch:

36:23 mark: looks like the Anagram will get the ability to do "in the box" profiling / capturing.

 

[Dom Mcsweeney]

36:23 mark: looks like the Anagram will get the ability to do "in the box" profiling / capturing.

Yeah, sounded like that to me too.

 

[Lysander]

36:23 mark: looks like the Anagram will get the ability to do "in the box" profiling / capturing.

If Anagram is able to build NAM profiles in-box i'll be mighty impressed.

(it wont)

 

[AFKAEjay(retired)]

I really like the formfactor of this unit., Nice I/O set also.
If there is proof of the pudding of the time efx, capability to split into cabsimmed/non cabsimmed, and utility stuff like global EQs and midi integration…I might consider it.

 

[2dor]

If Anagram is able to build NAM profiles in-box i'll be mighty impressed.

(it wont)

I do wonder how well it'll fare.

What I suspect they might do is work on some integration with the Tone3000 guys to allow "further training" a profile created on the Anagram but in the cloud.

I say this for 2 reasons:

- Steve recently released trainer version 0.12.0 this weekend (https://www.neuralampmodeler.com/post/trainer-v0-12-0-is-released) where he added the option to resume training from a .nam file (provided you have the input.wav and output.wav signals as well); this feature was an "under the hood" thing in past releases and it involved some hacky edits to resume training from a specific checkpoint

- The Darkglass guys were very specific in how they want the Anagram to become a platform & not necesarily just a one-trick pony. With Steve being now full-time on-board with NAM, I think he might be more involved in these quality of life enhancements that could further help Darkglass' profiling initiative.

 

[Humbug]

36:23 mark: looks like the Anagram will get the ability to do "in the box" profiling / capturing.

I don't like this. The whole point of learning how to make NAM-models was so I could feel superior to everyone else! It's also why I prefer to use Linux at home.

Seriously, though; if they get this working (relatively) painlessly it's a pretty good selling point.

 

[mhelin]

I don't like this. The whole point of learning how to make NAM-models was so I could feel superior to everyone else! It's also why I prefer to use Linux at home.

Seriously, though; if they get this working (relatively) painlessly it's a pretty good selling point.

I guess they will use the tone-3000 API from the device (meaning it must have online connection to internet using your phone for an example):

GitHub - tone-3000/api: TONE3000 API Integration Guide

TONE3000 API Integration Guide. Contribute to tone-3000/api development by creating an account on GitHub.

github.com

 

[mhelin]

I don't like this. The whole point of learning how to make NAM-models was so I could feel superior to everyone else! It's also why I prefer to use Linux at home.

Seriously, though; if they get this working (relatively) painlessly it's a pretty good selling point.

The CPU in Anagram is a Rasperry PI 5 class device with two extra Cortex-A55 CPU cores plus the NPU (neural prosessing unit). Wonder if the NPU is used a lot yet, it can perform at 6 TOPS at INT8 precision. Considering utilizing it Rockchip claims that "Users can easily convert network models based on frameworks such as TensorFlow/MXNet/PyTorch/Caffe, and can use RKNN and RKLLM frameworks for accelerated model inference." So they only mention inference (using the model) not modeling or training itself. You are not going to use python for inference anyway, but I think they offer some kind of RKNN API in form of shared object (.so) library for C/C++ code base (which must be the same one which is accessed from python code anyway). Anyway, the training using the RKNN and NPU might be possible but possibly quite slow (if the model fits into the memory of the NPU to start with). RK3582 based boards can be found from Aliexpress (made by Radxa) for less that 50€.

 

[2dor]

Been catching up on this thing and their last Facebook post was about Bluetooth functionality being released in aprox 1 week.

That said, the unit's been unavailable almost since it was "officially" launched & some folks on FB who preordered are pi553d A.F. since Darkglass announced they're only starting production on August 1st.

That means the unit will probably only be (properly) in-stock sometime in autumn; sucks big time for anybdoy that's been waiting for it. Happy I didn't put down an order & decided to let things unfold a bit.

 

[[Nathan]]

Been catching up on this thing and their last Facebook post was about Bluetooth functionality being released in aprox 1 week.

That said, the unit's been unavailable almost since it was "officially" launched & some folks on FB who preordered are pi553d A.F. since Darkglass announced they're only starting production on August 1st.

That means the unit will probably only be (properly) in-stock sometime in autumn; sucks big time for anybdoy that's been waiting for it. Happy I didn't put down an order & decided to let things unfold a bit.

That’s so weird with availability. I’d been getting ads from our local AU stockist. I just checked and yep no actual stock “taking orders”. I’d be pretty annoyed, totally takes the buzz out of it when this happens.

 

[2dor]

That’s so weird with availability. I’d been getting ads from our local AU stockist. I just checked and yep no actual stock “taking orders”. I’d be pretty annoyed, totally takes the buzz out of it when this happens.

Everybody wants to be nVidia nowadays - sheesh

 

[dronerstone]

Everybody wants to be nVidia nowadays - sheesh

They should ignore the stocks performance and rather see the failed launches and crazy pricing schemes as a motivation to not be like Nvidia.

 

[[Nathan]]

Everybody wants to be nVidia nowadays - sheesh

If someone gets NAM running on a switch2 it might tip me over the line