OBS Studio Noise Suppression: RNNoise vs Speex
OBS Studio offers two primary software-based noise suppression filters to clean up background audio: Speex and RNNoise. While both aim to eliminate unwanted ambient sounds like PC fans, keyboard clicks, or air conditioning, they use fundamentally different technologies to achieve this. This article explains the functional differences between Speex and RNNoise, helping you choose the best option for your streaming or recording setup based on CPU usage, audio quality, and configuration needs.
Speex: Low CPU Usage (Algorithmic)
Speex is a traditional, algorithmic noise suppression method. It works by analyzing the frequency spectrum of your audio and subtracting frequencies that match a steady-state noise profile.
- Manual Configuration: Speex requires manual adjustment. It provides a slider measured in decibels (dB) that allows you to set the suppression threshold (typically ranging from -30 dB to 0 dB). You must manually tune this slider to find the balance between blocking background noise and keeping your voice clear.
- Performance: It is extremely lightweight and consumes virtually no CPU resources. This makes it ideal for older computers, budget laptops, or systems running highly demanding games where every frame counts.
- Limitations: Speex is designed for static, continuous background noise (like a steady hum). It struggles with dynamic noises like keyboard clicks, barking dogs, or sudden movements. If the suppression level is set too high, it can make your voice sound metallic, robotic, or muffled.
RNNoise: High Quality (AI-Driven)
RNNoise is a modern, deep-learning-based noise suppression filter. It utilizes a recurrent neural network (RNN) trained on thousands of hours of clean voice and noisy audio samples to distinguish between human speech and background noise in real-time.
- Automatic Operation: Unlike Speex, RNNoise requires no manual tuning or sliders. It automatically detects what is speech and what is noise, muting everything that does not match the profile of a human voice.
- Audio Quality: RNNoise offers significantly better audio quality and suppression capabilities. It is highly effective at removing both static noises (like fans) and non-static, unpredictable noises (like mechanical keyboard typing, paper rustling, and mouse clicks) without distorting your voice.
- Performance: Because it uses an AI model, RNNoise requires more processing power than Speex. While modern CPUs handle this easily, it can occasionally cause audio stuttering or increased latency on low-end or heavily taxed hardware.
Quick Comparison
| Feature | Speex | RNNoise |
|---|---|---|
| Technology | Algorithmic (Spectral Subtraction) | AI / Recurrent Neural Network |
| Control | Manual (slider adjustment required) | Fully Automatic (no configuration) |
| CPU Usage | Negligible (extremely low) | Moderate |
| Best For | Lower-end PCs, steady hums | Modern PCs, dynamic noises, best voice quality |
| Handling of Clicks/Claps | Poor | Excellent |
Which One Should You Choose?
For the vast majority of creators, RNNoise is the superior choice because it provides much cleaner audio and handles real-world background noises like keyboard typing automatically. However, if you notice audio lag, robotic stuttering, or high CPU usage warnings in OBS, switching to Speex and manually adjusting the slider is the best way to reclaim system resources while still maintaining basic noise reduction.