From nobody Sun Jul 28 19:21:00 2024 X-Original-To: freebsd-multimedia@mlmmj.nyi.freebsd.org Received: from mx1.freebsd.org (mx1.freebsd.org [IPv6:2610:1c1:1:606c::19:1]) by mlmmj.nyi.freebsd.org (Postfix) with ESMTP id 4WXBDR5Mfmz5RV5G for ; Sun, 28 Jul 2024 19:21:11 +0000 (UTC) (envelope-from meka@tilda.center) Received: from chubby.tilda.center (meka.rs [109.93.255.137]) (using TLSv1.3 with cipher TLS_AES_256_GCM_SHA384 (256/256 bits) key-exchange X25519 server-signature RSA-PSS (4096 bits) server-digest SHA256) (Client did not present a certificate) by mx1.freebsd.org (Postfix) with ESMTPS id 4WXBDQ5VFZz45H5 for ; Sun, 28 Jul 2024 19:21:10 +0000 (UTC) (envelope-from meka@tilda.center) Authentication-Results: mx1.freebsd.org; dkim=none; dmarc=pass (policy=reject) header.from=tilda.center; spf=pass (mx1.freebsd.org: domain of meka@tilda.center designates 109.93.255.137 as permitted sender) smtp.mailfrom=meka@tilda.center Received: from [192.168.111.162] (hal9000.meka.rs [192.168.111.162]) by chubby.tilda.center (Postfix) with ESMTPSA id 51746961A2 for ; Sun, 28 Jul 2024 19:21:00 +0000 (UTC) Content-Type: multipart/alternative; boundary="------------QSrxY1Br1HB8BaRXtk1s4ATD" Message-ID: <36aa8e3c-cd44-4836-ba7a-5428402680e8@tilda.center> Date: Sun, 28 Jul 2024 21:21:00 +0200 List-Id: Multimedia discussions List-Archive: https://lists.freebsd.org/archives/freebsd-multimedia List-Help: List-Post: List-Subscribe: List-Unsubscribe: Sender: owner-freebsd-multimedia@FreeBSD.org MIME-Version: 1.0 User-Agent: Mozilla Thunderbird Subject: Re: Simulating guitar stack To: freebsd-multimedia@freebsd.org References: <81967489-dd50-4ace-bdad-88100c7cfc0c@tilda.center> <5853048.8T7jmnknE8@x230> Content-Language: en-US From: =?UTF-8?Q?Goran_Meki=C4=87?= In-Reply-To: <5853048.8T7jmnknE8@x230> X-Spamd-Bar: -- X-Spamd-Result: default: False [-2.95 / 15.00]; NEURAL_HAM_SHORT(-1.00)[-0.999]; NEURAL_HAM_LONG(-1.00)[-0.997]; NEURAL_HAM_MEDIUM(-0.98)[-0.980]; R_MIXED_CHARSET(0.71)[subject]; DMARC_POLICY_ALLOW(-0.50)[tilda.center,reject]; R_SPF_ALLOW(-0.20)[+mx]; ONCE_RECEIVED(0.10)[]; MIME_GOOD(-0.10)[multipart/alternative,text/plain]; XM_UA_NO_VERSION(0.01)[]; RCVD_VIA_SMTP_AUTH(0.00)[]; RCPT_COUNT_ONE(0.00)[1]; RCVD_COUNT_ONE(0.00)[1]; ASN(0.00)[asn:8400, ipnet:109.92.0.0/15, country:RS]; MIME_TRACE(0.00)[0:+,1:+,2:~]; RCVD_TLS_ALL(0.00)[]; R_DKIM_NA(0.00)[]; MLMMJ_DEST(0.00)[freebsd-multimedia@freebsd.org]; FROM_EQ_ENVFROM(0.00)[]; FROM_HAS_DN(0.00)[]; MID_RHS_MATCH_FROM(0.00)[]; TO_DN_NONE(0.00)[]; PREVIOUSLY_DELIVERED(0.00)[freebsd-multimedia@freebsd.org]; TO_MATCH_ENVRCPT_ALL(0.00)[]; ARC_NA(0.00)[] X-Rspamd-Queue-Id: 4WXBDQ5VFZz45H5 This is a multi-part message in MIME format. --------------QSrxY1Br1HB8BaRXtk1s4ATD Content-Type: text/plain; charset=UTF-8; format=flowed Content-Transfer-Encoding: 7bit On 4/30/24 23:16, Florian Walpen wrote: > I did experiment some years ago. That didn't include neural amp models back > then, but I just had a listen to some current NAM demos and my verdict is > still the same: > > - Most pedals can be easily reproduced in software. > - Nothing in software sounds like a real tube amp (yet). > - Speakers can be reproduced well by IR software. > > Means if you really care about tube amp tone, use a tube amp or at least a > hardware amp modeler that utilizes tubes (I have an old one made by VOX). That > said, many people will not hear the difference in a complete mix, and for > heavily distorted sounds it's less relevant. > > I personally don't use software effect pedals. For Speaker simulation I > recommend the lsp-plugins IR processor, and I apply some older free IRs from > Kalthallen Cabs. I experimented a lot with IR lately as I found that my FX processor for live gigs allows me to upload custom ones. In studio, lsp-plugins works great! It has everything I expect IR plugin to have. Thank you for the tip! As for neural amp modeling, I manually compiled following plugins: * https://github.com/mikeoliphant/neural-amp-modeler-lv2 * https://github.com/AidaDSP/aidadsp-lv2 They both work but first one has only one preset (hopefully others will follow as time goes by), and second one is not much better: https://tonehunt.org/models?tags%5B0%5D=aida-x. From such a limited experience I would say the technology is promising. There's a lot more to be done in this field to make it really shine, but what I heard I already like more than guitarix and/or rakarrack. As for capture, I do have problems. Luckily I found aliki from https://kokkinizita.linuxaudio.org/linuxaudio/ and managed to compile it and run it, but not yet capture any IR (I have a clue what's the problem, I'll write if I get stuck). Capturing for neural amp is trickier. No matter how I record it with ardour I get different number of samples for input and output file. I tried appending silence to both files (well, tracks really, as it's ardour I'm using for capturing) and aligning them, listening to what jack_iodelay tells me, reseting latencies ... everything I could think of, but in and out file always have different number of samples. My reamping works great and I can hear my gear just fine, but managing to produce the proper file(s) is tricky. The procedures I followed: * neural amp modeler: https://colab.research.google.com/github/sdatkinson/NAMTrainerColab/blob/main/notebook.ipynb * aida: https://colab.research.google.com/github/AidaDSP/Automated-GuitarAmpModelling/blob/aidadsp_devel/AIDA_X_Model_Trainer.ipynb (it has some errors currently, but reports different sample sizes) I stream-exported ardour tracks with "apply track processing" off, otherwise all files are saved as stereo. If anyone can recommend how to capture the sound correctly for training any of the two AI implementations, I would be really grateful. I tried to record my studio FX processor which introduces additional latency, but I also set latencies according to jack_iodelay. Please advise if you have any ideas! Regards, meka --------------QSrxY1Br1HB8BaRXtk1s4ATD Content-Type: text/html; charset=UTF-8 Content-Transfer-Encoding: 7bit
On 4/30/24 23:16, Florian Walpen wrote:
I did experiment some years ago. That didn't include neural amp models back 
then, but I just had a listen to some current NAM demos and my verdict is 
still the same:

 - Most pedals can be easily reproduced in software.
 - Nothing in software sounds like a real tube amp (yet).
 - Speakers can be reproduced well by IR software.

Means if you really care about tube amp tone, use a tube amp or at least a 
hardware amp modeler that utilizes tubes (I have an old one made by VOX). That 
said, many people will not hear the difference in a complete mix, and for 
heavily distorted sounds it's less relevant.

I personally don't use software effect pedals. For Speaker simulation I 
recommend the lsp-plugins IR processor, and I apply some older free IRs from 
Kalthallen Cabs.

I experimented a lot with IR lately as I found that my FX processor for live gigs allows me to upload custom ones. In studio, lsp-plugins works great! It has everything I expect IR plugin to have. Thank you for the tip!

As for neural amp modeling, I manually compiled following plugins:
* https://github.com/mikeoliphant/neural-amp-modeler-lv2
* https://github.com/AidaDSP/aidadsp-lv2

They both work but first one has only one preset (hopefully others will follow as time goes by), and second one is not much better: https://tonehunt.org/models?tags%5B0%5D=aida-x. From such a limited experience I would say the technology is promising. There's a lot more to be done in this field to make it really shine, but what I heard I already like more than guitarix and/or rakarrack.

As for capture, I do have problems. Luckily I found aliki from https://kokkinizita.linuxaudio.org/linuxaudio/ and managed to compile it and run it, but not yet capture any IR (I have a clue what's the problem, I'll write if I get stuck). Capturing for neural amp is trickier. No matter how I record it with ardour I get different number of samples for input and output file. I tried appending silence to both files (well, tracks really, as it's ardour I'm using for capturing) and aligning them, listening to what jack_iodelay tells me, reseting latencies ... everything I could think of, but in and out file always have different number of samples. My reamping works great and I can hear my gear just fine, but managing to produce the proper file(s) is tricky. The procedures I followed:
* neural amp modeler: https://colab.research.google.com/github/sdatkinson/NAMTrainerColab/blob/main/notebook.ipynb
* aida: https://colab.research.google.com/github/AidaDSP/Automated-GuitarAmpModelling/blob/aidadsp_devel/AIDA_X_Model_Trainer.ipynb (it has some errors currently, but reports different sample sizes)

I stream-exported ardour tracks with "apply track processing" off, otherwise all files are saved as stereo. If anyone can recommend how to capture the sound correctly for training any of the two AI implementations, I would be really grateful. I tried to record my studio FX processor which introduces additional latency, but I also set latencies according to jack_iodelay. Please advise if you have any ideas!

Regards,
meka

--------------QSrxY1Br1HB8BaRXtk1s4ATD--