Quote:
Ah .ok .. i notice few more things:
- Voices are recovered veru good and peeks from recovered voices are high, more then from mid-lows and lows.
- Decliping is very depend of Input level! So, "internal" Double AGCing may fix this. Double: fiist on input and second on output but Vice versa.. to maintain always same level.. internal: inside only for filter functioning.
- That makes sense - voices usually have much higher spikes than other frequencies
- I'm going to add an option to dynamically detect the clipping value. (For hard clipping, that's easy - the highest value available in the audio. But for MP3 encoded or tilt corrected audio I need to think of something a bit smarter.)
When that works I'll also try to let the tilt be determined automatically. (That should be easy: Always try current tilt, tilt plus a bit and tilt minus a bit, and move towards the tilt value which gives the steepest histogram ending.)