- Useful only when it improves the signal that reaches VAD and STT. An aggressive filter can remove consonants or distort speaker cues — cleaner-sounding audio is not always better recognition input.
- RTF matters because denoising sits on every incoming frame. A model that is accurate but eats the latency budget is the wrong front-end model.
Connections
- measured_by Real-Time Factor
- precedes Voice Activity Detection