It is important to note that librnnoisevstdll is strictly a Windows file. If you are on macOS or Linux, you would be looking for a .vst or .so file, respectively. Additionally, older DAWs might require a "bridged" version if the DLL is 64-bit and your software is 32-bit (or vice versa), though most modern setups handle this automatically.
Windows audio plugins generally live in specific folders. If you don't have one, create a folder named VST Plugins on your hard drive (e.g., C:\Program Files\VST Plugins).
RNNoise is a noise suppression library based on a recurrent neural network (specifically a GRU - Gated Recurrent Unit). Unlike traditional noise gates that simply cut audio below a volume threshold, or standard noise reduction that uses spectral subtraction, RNNoise is trained to recognize the difference between human speech and noise.
#include "rnnoise.h" #include <stdio.h>int main() DenoiseState *st = rnnoise_create(NULL); if (!st) printf("Failed init\n"); return 1;
float noisy[480] = 0; float clean[480]; // ... fill noisy with 48 kHz mono audio ... float vad = rnnoise_process_frame(st, clean, noisy); printf("VAD probability: %f\n", vad); rnnoise_destroy(st); return 0;
Compile (MSVC example):
cl /I. /Fe:test.exe test.c rnnoise.lib
(Place rnnoise.dll in same dir.)
The file librnnoisevst.dll (often part of RNNoise-VST) is a real-time noise suppression plugin based on the RNNoise library developed by Xiph.Org. It uses a Recurrent Neural Network (RNN) to distinguish human speech from background noise. Performance Review
Noise Removal: Highly effective at removing static background noise such as computer fans, office hum, and air conditioning. It generally outperforms traditional noise gates because it filters noise while you are speaking, rather than just muting the mic during silence.
Audio Quality: While it is "small and fast," it can sometimes produce a "robotic" or "metallic" quality to the voice, especially if the input signal is weak or the noise is extremely loud.
Latency: Users have reported mixed results regarding delay. While the underlying library is designed for real-time use, some VST implementations may introduce noticeable latency depending on the host application (like OBS or Equalizer APO).
Resource Usage: Extremely lightweight on CPU and does not require a GPU, making it a popular alternative to hardware-heavy solutions like NVIDIA Broadcast. Critical Technical Requirements
To avoid common bugs (like no audio or heavy distortion), you must ensure your audio settings match the plugin's fixed requirements:
Sample Rate: The plugin typically only works at 48,000 Hz (48 kHz). Using other rates like 44.1 kHz often results in silence or severe distortion. Format: It generally expects 16-bit audio input. Popular Implementations librnnoisevstdll
librnnoisevst.dll (often referred to as the RNNoise VST) is an open-source, AI-powered noise suppression plugin based on the RNNoise library
. It uses a Recurrent Neural Network (RNN) to differentiate human speech from background noise in real-time without requiring high-end hardware like GPUs. Key Features and Capabilities AI-Driven Suppression
: Unlike traditional "noise gates" that simply mute audio below a certain volume, this plugin actively identifies and filters out specific sounds like computer fans, traffic, and mechanical keyboard clicks. Efficiency
: It is designed to be lightweight, running entirely on the CPU with very low impact on system performance. Real-Time Processing : Ideal for live scenarios such as streaming on OBS Studio or conferencing. Voice Activity Detection (VAD)
: The plugin includes a VAD threshold to ensure that only voice signals pass through, silencing other sounds when you aren't speaking. Common Use Cases VST noise-gate / expander free plugins - Wavosaur
To produce a "proper piece" using librnnoise_vst.dll (often referred to as the Werman RNNoise VST plugin), you need to correctly integrate it into your audio chain. This plugin uses a Recurrent Neural Network (RNN) specifically trained to isolate human speech from background noise. 1. Installation & Placement
Download & Extraction: Download the latest release (32-bit or 64-bit to match your OS) and unzip it. It is important to note that librnnoisevstdll is
File Path: Copy librnnoise_vst.dll to a stable location, such as C:\Program Files\VSTPlugins or a direct folder on your C: drive to avoid access permissions.
System Sample Rate: For optimal performance, set your microphone and playback devices to 16-bit, 48,000 Hz (48 kHz) in Windows Sound Settings. 2. Integration with Host Software
Because the DLL is a VST plugin, it requires a "host" to run. Common setups include: OBS Studio: Right-click your Mic/Aux source and select Filters. Add a VST 2.x Plug-in filter. Select librnnoise_vst.dll from the dropdown list.
Note: OBS has a built-in "RNNoise" option, but the VST version allows for more specific routing if needed. Equalizer APO (System-wide): Open Editor.exe and add a new VST Plugin block. Navigate to your librnnoise_vst.dll file. This applies the noise suppression to all Windows audio. 3. Fine-Tuning for a "Proper" Output
Mono vs. Stereo: Use rnnoise-stereo.dll if you are working with stereo sources, but most microphone setups perform better with the mono-focused version to avoid "one-ear" audio issues.
Placement in Chain: Always place RNNoise first in your filter chain. Suppressing noise before adding compression or EQ ensures the AI isn't trying to process amplified background hiss.
Avoid Over-Processing: High-intensity screaming or laughing can sometimes confuse the AI, leading to robotic or "watery" artifacts. Compile (MSVC example): cl /I
Are you setting this up for live streaming in OBS or for system-wide use in professional calls?