Deep Vintage

Legendary Hardware Simulations with a Deep Vintage Soul

Deep Vintage is a suite of legendary hardware simulation plugins that will immerse you in real analog magic. Not merely pursuing bona fide circuit replication or specific tonal qualities, Deep Vintage simulates the entirety of sound to reproduce the vintage “soul” in a digital world. Depth, sheen, low-end saturation…every nuance of the hardware’s sonic spirit is ready for you to ignite, with minimal CPU usage and latency.

APNN 2.0

Trained with Three-Body Tech’s proprietary APNN (Audio Processing Neural Network) 2.0, Deep Vintage learns on sound, and sound alone, creating an indistinguishable listening experience from the original hardwares.

APNN 2.0 is a neural network specializes in simulating analog hardwares. During the training process, APNN 2.0 and the hardware will be inputted with the same audio, and APNN 2.0 will learn how the hardware changes the audio in both waveform and spectrum dimensions. This means that a well-trained APNN 2.0 instance can capture both the dynamic and tonal characteristics of its source hardware. The following diagram demonstrates how, as training progresses, APNN 2.0’s waveform and spectrum response deviation gradually decrease, eventually becoming indistinguishable from that of the original hardware. Check out the corresponding demos below to hear how APNN 2.0 progressively learns and replicates the sound of the hardware.

2000th Iteration

12000th Iteration

70000th Iteration

Final

Real Hardware

Loading...

How is it different?

We have been exposed to numerous emulation technologies: physical modeling, convolution... among many others. Regardless of what technics to engage, our ultimate goal is consistent: to achieve the highest possible fidelity in reproducing the sound of the real-world devices.

APNN was meticulously designed and trained to excel in this task, as its learning is based on audio signals from the real hardware units. Deep Vintage plugins are not merely 'theoretically correct modeling of circuits' nor solely based on harmonic structures or impulse responses; they capture 100% 3D feels in the analog sound.

How does it work?

To put it simply, APNN is a neural network specifically optimized for audio processing, which is composed of a massive combination of audio processing modules (EQ, compressor, overdrive, etc.).

To capture the essence of a hardware model, APNN will automatically adjust its structure and parameters until the difference between its output and the hardware's output progressively diminishes.

Ultimately, APNN achieves an error signal of phase cancellation at about -40dB to -75dB (depending on the hardware model). With this exceptional level of error control, which surpasses even the variance between different production batches of the same hardware model, we can confidently say that APNN is capable of 'deceiving' human ears.

After an APNN 2.0 instance completes its training, we conduct rigorous human testing and make adjustments until our entire team fails the ABX test. This allows us to proudly announce:

in the realm of digital audio, nothing comes closer to real hardware than Deep Vintage.

Highlights

Multiplex Saturation, Multistage Coloration

With the power of APNN 2.0, Deep Vintage simulates not just specific frequency responses or coloration, but all the subtle 'hardware mojos': dynamics, airness, phase shifts, tube voltage sag, transformer’s “iron sound,“ and more. Whether for subtle coloration, moderate saturation, or crushing the entire audio, its authentic performance will make you forget it's digital.

Independent Harmonics Control

With real hardware, the amount of harmonics is fixed at a given knob setting. However, Deep Vintage introduces surreal flexibility by allowing independent control over harmonics, separate from all other tonal characteristics. This lets you dial in the sonic power of high drive settings while maintaining the purity of a clean tone.

Low Frequency Saturation

The 'iron' sound of audio transformers—gently added low-end girth and saturation—epitomizes the sonic character of real hardware. Deep Vintage not only accurately captures this, but also provides you with the ability to toggle this 'iron' sound on or off, allowing you to switch between transformer or transformer-less versions. Whether you're aiming for a thick or clear tone, it always delivers with exceptional quality.

Re-sampling/Up-sampling

Almost all audio processing networks operates in fixed sample rates, but we’ve made resampling possible by optimizing our networks. The completely redesigned resampling algorithm in Deep Vintage ensures consistent accuracy and fidelity across all sample rates, making the simulation fully sample rate agnostic. Additionally, up to 8x oversampling is supported, effectively eliminating any aliasing issues.

Harmonic distortions with oversampling disabled, resulting in heavy aliasing artifacts at high frequencies.

Harmonic distortions with 4x oversampling enabled, showing no aliasing artifacts

EQ Co-training

Most neural networks can only capture discrete states of the hardware, thus providing only limited EQ combinations. However, Deep Vintage uniquely supports fully continous EQ adjustment through extra EQ simulations. For models with EQ, the "co-training algorithm" simultaneously learns the hardware prototype's saturation characteristics while fine-tuning a pre-modeled EQ module based on the circuit. This allows you to enjoy the authentic hardware sound while having complete freedom over EQ adjustment.

Tape Wow/Flutter Co-training

Wow/Flutter are pitch variations that occur in tape machines due to mechanical inconsistencies in the tape transport system. Wow refers to slower, more noticeable pitch fluctuations, while flutter on the other hand, is a faster form of speed variation.

Just like EQ co-training, APNN 2.0 uses a physically modeled wow/flutter simulation and co-trains it with the neural network. This not only makes the results of neural network training sound more authentic but also brings the modeled wow/flutter effect closer to the original hardware.

Adjustable Noise Floor

The Deep Vintage series simulates the hardware’s inherent noise floor, which you can adjust in amount as needed.

"Retro DAW" Simulation

This button is inspired by the extremely subtle changes (less than -140 dB) introduced by a classic DAW from around the year 2000. While the changes are infinitesimal, we didn’t overlook them. You can enable or disable this feature as needed.

Low CPU usage

No need for expensive cloud-based GPU clusters—Deep Vintage runs locally just like any other plugin, with extremely low CPU usage. You can easily insert it on every track!