#492 Tienes un post de Micky donde lo cuenta todo aquí
https://www.guitarristas.info/reportajes/era-nam-fabricantes-ponen-acuerdo-remedio/12106
Una IR acostumbra a ser la captura de una cabina+altavoz+micro+previo de micro.
Cuando no es eso, son capturas de reverbs/espacios (convolution reverbs).
NAM captura amplis, compresores, etc.
#494 Conceptualmente es lo mismo pero está hecho de una forma radicalmente distinta.
Una captura de Kemper está basada en un modelado base a la que se le impone, por así decirlo, una ecualización.
Kemper detecta que el ampli es "tipo fender" y sobre su modelo fender, le añade características para que se parezca en sonido y respuesta al ampli que se captura.
Una captura NAM está haciendo algo más parecido a una IA generativa a base de ir comparando un sonido base con un sonido grabado que subes tú a un servidor.
Alguien escribió:
How does NAM work?
Neural Amp Modeler works by training an neural network to learn the input-output relationship of an amplifier as a mathematical function. Instead of simulating the physics of tubes, transformers, capacitors, and circuits, NAM analyzes the actual behavior of real gear by studying what audio comes out when specific signals go in. Here's the NAM capture process:
Sweep Signal: A specially designed training signal called a "sweep signal" is sent through the gear you want to model
Recording: Both the input signal (DI) and the output from the mic'd amp are recorded simultaneously at high quality
Training: A neural network is trained on these paired recordings, learning to predict the output for any given input. This typically takes 5-20 minutes depending on model complexity.
Validation: The trained model is tested against held-out audio to verify accuracy
What does training a NAM model actually do?
The key advantage of NAM is its ability to capture complex nonlinear distortion, compression, and saturation characteristics that make amplifiers sound unique and respond naturally to playing dynamics.
How does NAM work?
Neural Amp Modeler works by training an neural network to learn the input-output relationship of an amplifier as a mathematical function. Instead of simulating the physics of tubes, transformers, capacitors, and circuits, NAM analyzes the actual behavior of real gear by studying what audio comes out when specific signals go in. Here's the NAM capture process:
Sweep Signal: A specially designed training signal called a "sweep signal" is sent through the gear you want to model
Recording: Both the input signal (DI) and the output from the mic'd amp are recorded simultaneously at high quality
Training: A neural network is trained on these paired recordings, learning to predict the output for any given input. This typically takes 5-20 minutes depending on model complexity.
Validation: The trained model is tested against held-out audio to verify accuracy
NAM Flow Chart
Diagram of Neural Amp Modeler capture process.
What does training a NAM model actually do?
The key advantage of NAM is its ability to capture complex nonlinear distortion, compression, and saturation characteristics that make amplifiers sound unique and respond naturally to playing dynamics.
https://www.tone3000.com/guides/neural-amp-modeler#how-does-nam-work
