Hi, after the aliasing noise topic on Fractal forums, Neural DSP Facebook group for the Quad Cortex and the paper from the creators of the capture (machine learning with neural network, link to follow) I decided to run some tests.
First a free plugin with up to 8x internal oversampling option to avoid aliasing.
(Kemper goes to 16x if I remember, I never could spot aliasing, same for AxeFxIII Mark 1, will upload the test if you ask)
Second is Neural DSP Capture tech inside Gojira Plugin. Neural DSP claimed they switched entirely in a black box style modeling for this plugin and so I guess using the same capture technology inside quad Cortex. No option for internal oversampling.
Be aware internal oversampling is not the sample rate of your project. The most important is the internal oversampling, the rest will be very low inaudible aliasing, at 44,1KHz or 96, very few people can spot the difference.
Understanding what is exactly Aliasing will need some scientific knowledge BUT it’s never too late and here’s some links to go deeper and understand.
I always found Neural DSP a little harsh sounding and ear tiring. But never exactly knew why before all the discussions and testing.
Basically Aliasing occurs when you use digital distortion or gain stages (but also with non linear treatments) and it generates artifacts that can be eared. Keep in mind that this test is very simple, with a sweeping sine wave generated, so in real life with a guitar pickup all of this is a lot more problematic, and this noise go all around the spectrum and it’s not harmonically related, so it’s purely unwanted noise.
Pure analog distortion don’t aliase at all so the sound is « clearer » and more precise even with huge amount of distortion.
Thanks for watching, and thanks to Jean to have come to me to ask for this test!
Began as a very agressive conversation but ended up with constructive conversation and great positive thoughts as the goal is common for all of us customers = the Tone.
Don’t be fooled by marketing, always follow your ears!
For aliasing better understanding =
• Samplerates: the highe...
Neural Network and machine learning of guitar amp Scientific Paper, some authors working now for Neural DSP and neural capture =
www.mdpi.com/2076-3417/10/3/7...
Neural DSP about Gojira = "A new audio engine powering Neural DSP pluginsOur analog modeling technology has been shifting towards completely automated machine-learning-based methods.We have developed a new audio engine that uses a proprietary anti-derivative trigonometric interpolation (AATI) algorithm to convert any sample rate used in the system or DAW to an optimum internal processing sample rate. The AATI algorithm allows us to use our machine-learning-based black-box modeling techniques for a wide range of effects in special amplifiers and distortion pedals.The benefit of using our AATI algorithm is vastly improved stability and CPU efficiency, allowing us to achieve optimal CPU performance and audio quality for any DAW sample rate. This change removes the need for an oversampling switch in our plugins - a change many users have noticed in Archetype: Gojira, with which many users are reporting dramatically improved CPU performance.We intend to roll this change out to all our other plugins over the following months. - Neural DSP"
31 янв 2021
