百科.dev
全部条目在线工具AI 编程趋势榜开源项目技术资讯提交条目
登录
< 返回工具列表
P

Proteus

> 编程语言
开源

使用神经网络的吉他放大器和踏板捕获插件。

198 stars0 点赞2 次浏览
访问官网GitHub

工具介绍

使用神经网络的吉他放大器和踏板捕获插件。

Proteus

Capture your own amps/pedals/plugins with Proteus. Can capture a drive/tone knob, or snapshot of the sound at a specific setting. Use the Proteus Capture Utility to quickly train models in the cloud with Colab. Effective for Amps/PreAmps, Distortion/Overdrive/Boost pedals (non-time based, no Reverb/Delay/Flange/Phaser). You can also capture a "rig", or combination of pedals/amp. This is similar in concept to a Kemper, Quad Cortex, or ToneX, in a free and open source plugin, with the ability to capture and share the sound of guitar gear normally costing hundreds or thousands of dollars.

Go directly to Colab to start training models.

  • Checkout the video tutorials for creating your own models for the Proteus plugin.

    • Amp Capture
    • Pedal Capture
    • Plugin Capture
  • Visit the GuitarML ToneLibrary Website to download Proteus compatible models.

  • Download the Proteus Capture Utility, which includes the input audio file and Colab script to train models for Proteus.

  • As of Version 1.2, Proteus also features a 3-Band EQ (Bass, Mid, Treble) and a built in default IR (for convenience, recommended to use a dedicated IR plugin).

Proteus uses a LSTM neural network to emulate guitar amplifiers/preamps and distortion/overdrive/boost pedals. You can capture the sound of an amplifier either by recording with a microphone, or direct out from a load box. When running "Direct Out" models, you will need to use an Impulse Response plugin to accurately model the amp speaker/cabinet.

You can create your own models using the Automated-GuitarAmpModelling repository directly (LSTM with hidden size 40), or by using the Capture Utility files (available for download at GuitarML.com) with Google Colab (recommended).

To share your best models, email the json files to [email protected] and they may be included in the ToneLibrary.

Installing the plugin

  1. Download the appropriate plugin installer (Windows, Mac, Linux) from GuitarML.com or the Releases page.
  2. Run the installer and follow the instructions. May need to reboot to allow your DAW to recognize the new plugin.
  3. Download the Proteus Capture Utility from GuitarML.com to create your own amp/pedal/plugin captures.

Loading models

Download available models from the Proteus ToneLibrary. Use the Load Model button to select a folder containing Proteus json models. Note that models for NeuralPi and SmartPedal use a different model architecture and will not be compatible.

Capturing an Amp/Pedal/Plugin

  1. Download the Proteus_Capture_Utility.zip to obtain the ProteusCapture.wav and ProteusCapture.ipynb Colab script.
  2. Play the ProteusCapture.wav from your DAW / Audio Device and input to your amp/pedal/plugin, and record the output. Recommended to use a Reamp for impedence matching, and a Load Box for direct amp captures.
  3. Upload the "ProteusCapture.ipynb" from the Capture Utility to the Google Colab website. Upload your out.wav file and follow the instructions in the Colab script to create a Proteus compatible model. Snapshot model training will take 10 minutes or less, and Knob model training will take around 30 minutes. For Knob captures, you will need 5 separate out wav files, as detailed in the tutorial videos.
  4. Download your model file (json format) and load into the Proteus plugin.

Note: Recommended to follow along with the video tutorials listed above.

Troubleshooting Captures

  1. The model training is very sensitive to any latency in your output recording. Audio timing mismatch will result in stalled training, where it never goes below 0.75 loss value. May be required to manually line up your audio files before exporting, using the initial click (approx. 1 second into the ProteusCapture.wav) as a guide.
  2. You can capture amps using a microphone, which will result in the capture of the mic/cab/speaker. These captures may be less accurate than direct amp captures from a load box. Direct amp captures are also more flexible for adding Impluse Responses after the Proteus plugin.
  3. Recommended to test a snapshot capture of your device before attempting a Knob capture. Snapshot captures are much quicker, less prone to error, and can be more accurate.
  4. When exporting your audio recording, it needs to be WAV format, PCM16 (FP32 also acceptable), 44100 samplerate, Mono.
  5. In Colab, the different devices listed for Step 4a,b,c are just guidelines. They start from a pre-trained model. You may be able to get better results by trying a different option for step 4.
  6. You may use different input audio for training, either by modifying the ProteusCapture.wav or using your own. If you do this, modify the training line in Colab step 4 in the following way (using your custom file for "YourNewInput.wav") and upload both wav files to Colab:

!python prep_wav.py $model -s ../YourNewInput.wav ../out.wav --normalize true 7. The sound you record is the sound the code will try to capture. Minimize noise/unwanted sounds for best results. You can also modify your recording with EQ or other effects.

Build Instructions

Build with Cmake

bash
# Clone the repository
$ git clone https://github.com/GuitarML/Proteus.git
$ cd Proteus

# initialize and set up submodules
$ git submodule update --init --recursive

# build with CMake
$ cmake -Bbuild
$ cmake --build build --config Release

The binaries will be located in Proteus/build/Proteus_artefacts/

Info

The neural network used in Proteus is a re-creation of the LSTM inference model from Real-Time Guitar Amplifier Emulation with Deep Learning

The Automated-GuitarAmpModelling project was used to train the .json models.

GuitarML maintains a fork with a few extra helpful features, including a Colab training script. IMPORTANT: When training models for Proteus, ensure that a LSTM size of 40 is used. Proteus is optimized to run models of this size, and other sizes are not compatible.

The plugin uses RTNeural, which is a highly optimized neural net inference engine intended for audio applications.

Special Thanks

Special thanks to John Stutts and Stefan Schmidt for the graphics used in Proteus.

Issues· 0 开放

查看全部 Issues在 GitHub 打开

暂无开放 Issues,或尚未同步最近议题。

> 标签

C++audio-processingguitarmachine-learningneural-networks

暂无评论,来聊聊你的看法吧

> 工具信息

发布日期2026年8月1日
最后更新2026年9月18日
分类编程语言
定价开源

> 相关工具

T
TypeScript
JavaScript 的超集,为前端与全栈提供静态类型
P
Python
通用编程语言,广泛用于 Web、数据与 AI
G
Go
Google 推出的简洁高效系统语言