DeepFilterNet causes crackling in heard audio only when actively speaking
System Information:
https://github.com/user-attachments/assets/eaa92c50-5a25-4afd-a6ee-45af88f34d6f
Problem: Audio output (headphone/speaker audio) has severe crackling and distortion only when I'm actively speaking with DeepFilterNet enabled on my microphone. Silent periods are fine - the issue only occurs during voice activity.
Symptoms:
- Crackling, popping, and stuttering sounds in headphone/speaker output
- Only happens when speaking into the microphone (voice activity)
- Silent moments have no crackling
- Affects all audio output (music, system sounds, etc.) when speaking
- Consistent across both Arch Linux and Fedora
Current Workaround:
Increasing buffer size (setting default.clock.min-quantum = 512 in PipeWire config) eliminates the crackling. However, I believe this underlying issue should be addressed in DeepFilterNet itself, as requiring system-wide configuration changes and accepting latency penalties is not an acceptable long-term solution.
Expected Behavior: DeepFilterNet should process microphone input without causing crackling in audio output, even at PipeWire's default quantum settings (32).
Reproduction:
- Install and enable DeepFilterNet LADSPA plugin for microphone input
- Use default PipeWire settings (min-quantum = 32)
- Play any audio (music, video, etc.)
- Speak into microphone
- Audio output crackles only during speech, not during silence
Additional Context: The issue suggests DeepFilterNet's processing load spikes during voice activity, causing buffer underruns that affect the entire audio pipeline with smaller quantum values.