Meanwhile, there's the nagging question of what to do with unique
An algorithm that can tell the difference between sound (coherent) and
noise (non-coherent) would be great. Or maybe a really sensitive noise
gate that can select pixel by pixel, or the audio equivalent. Imagine a
spectral view of a wav file and a process for eliminating everything
below a certain threshold on a pixel by pixel basis. Never mind the
slices of the spectrum. Only that which is above the threshold remains.
With the Berliner file in question, the voice fades somewhat at moments
leaving the noise as loud as the voice, or louder.
The problem for me is that I don't speak German so I don't what I'm
On 6/21/2012 11:34 AM, Don Cox wrote:
> On 21/06/2012, Randy Riddle wrote:
>> I've thought for some time that there's already a way to do this with
>> at least some recordings.
>> For years, film restorers have used multiple prints of films, taking
>> the best quality sections from each that survive, sometime
>> substituting small sections in a print that has been damaged.
>> Why couldn't that be done with recordings where multiple copies
>> Basically, what the software would do is let you take multiple sound
>> files sourced from different copies of the same record. Each will
>> have been damaged and degraded in different ways and have different
>> patterns of noise.
>> The software, after synching the recordings, would compare them and
>> "toss out" the noise and keeping commonalities between the copies.
>> The more copies of the recording you have available, the better the
>> result might be, at least theoretically.
>> Why couldn't this work?
> Wow from slight off-centring would be a problem, but maybe Capstan could
> deal with that.
> Otherwise, the stitching algorithms used for images should be easily
> The two groove walls of a mono record already do what you suggest, but
> twenty copies might be better than two.