[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [ARSCLIST] historical stylus rake angles (SRA)
On 08/01/05, Pete Tinker wrote:
> My (admittedly limited) experience has convinced me that no algorithms
> exist that will eliminate the noise. In this instance, noise has
> usually replaced the signal (the recorded sound), and any
> computational approach to restoring the signal would have to create
> "new" signal by trying to infer the original signal from existing
> signal surrounding the noise. Algorithms are typically only useful
> when you have to extract signal from a combination of signal+noise.
Which is what you have to do when playing an analog disc.
The human brain does this well - you know what is a click and what
is a muted trumpet. Writing a computer algorithm to distinguish clicks
and noises from music is difficult, but not impossible.
A useful approach is to look for a rate of change (ie the slope of the
waveform) which is much greater than the average for, say, the
surrounding quarter of a second.
It's no good just looking for spike-shaped waveforms, as (in particular)
a muted trumpet is a sequence of spikes each of which is the same shape
as a click. The difference is that in the case of the trumpet there are
many such spikes, whereas clicks from scratches or dust are isolated.
A higher sample rate will make it easier for the algorithm to
distinguish between clicks and recorded sounds.
Yes, once a click has been located, the original shape of the wave has
to be restored. IMO smoothing gives better results than interpolation.
Sometimes you can copy the wave from the other channel.