To some extent the metallic sound is an inherent artifact of the phase vocoder process. It is convolution smear. The best way I know to control this is to produce an analysis that consists of a large number (hundreds) of partial tracks, and resynthesize using an oscillator bank with a large number of oscillators. This seems to reduce the smearing, at the cost of simplifying the sound and requiring a lot of processor power. It may also be possible to apply further processing to the partial tracks to get rid of high-frequency ringing sounds, before resynthesis.

Others may have more experience in this area.

Hope this helps,
Mike

-----Original Message-----
>From: Hector Centeno 
>Sent: Oct 14, 2008 3:34 PM
>To: csound@lists.bath.ac.uk
>Subject: [Csnd] looking for advice on warming up phase vocoder processed sounds
>
>Hello,
>
>I've been using a lot the PVs family of opcodes to process sound
>samples (doing time stretches, pitch shifting, smoothing, etc.) and
>very often struggled trying to avoid "metallic" sounding results,
>specially when doing time stretches (i.e. reading a PVX file with a
>very slow pointer). One way I've been dealing with this is by using EQ
>and reverb (tried also pvsmooth and pvsblur and didn't help much), but
>I was wondering, are there any other techniques to avoid this? Is this
>the result of analyzing using wrong parameters (I usually do: pvanal
>-n 2048)? Do the PVS opcodes interpolate between windows? I've read in
>some online articles about FFT in MAX/MSP and found mention to
>interpolation as a way of increasing sound quality. Would this be a
>way of avoiding it?
>
>Thanks!
>
>Hector
>
>
>Send bugs reports to this list.
>To unsubscribe, send email sympa@lists.bath.ac.uk with body "unsubscribe csound"

Also the ATS opcodes might be useful. After analyzing a sound, it is 
represented in sine waves plus critical noise bands (as opposed to 
Fourier analysis, which use sinusoidals only). This allows for a 
stretching without the by-product of  PV (i.e. metallic sound). If I 
understand correctly, the metallic character of PV stretching comes from 
turning non-periodic information into periodic one (by means of 
stretching, or transposing). However, I'm not sure if it is possible to 
use the ATS opcodes in real time. I hope this is useful.
Cheers,

Ernesto

Hector Centeno wrote:
> aah, sorry, they seem to do it. It says so in the Csound manual entry
> for psfread:
>
> "Create an fsig stream by reading a selected channel from a PVOC-EX
> analysis file loaded into memory, with frame interpolation."
>
> Hector
>
>
>
> On Tue, Oct 14, 2008 at 3:57 PM, Hector Centeno  wrote:
>   
>> Mike,
>>
>> Thanks for your advice, it is helpful. I guess I should ad that I
>> would like to keep it CPU efficient so realtime synthesis can be done
>> (that's how I've been working).
>>
>> Do you (or anyone) know if the pvsfread and pvsdiskin interpolate when
>> reading from a table/file?
>>
>> Cheers,
>>
>> Hector
>>
>>
>> On Tue, Oct 14, 2008 at 3:46 PM, Michael Gogins  wrote:
>>     
>>> To some extent the metallic sound is an inherent artifact of the phase vocoder process. It is convolution smear. The best way I know to control this is to produce an analysis that consists of a large number (hundreds) of partial tracks, and resynthesize using an oscillator bank with a large number of oscillators. This seems to reduce the smearing, at the cost of simplifying the sound and requiring a lot of processor power. It may also be possible to apply further processing to the partial tracks to get rid of high-frequency ringing sounds, before resynthesis.
>>>
>>> Others may have more experience in this area.
>>>
>>> Hope this helps,
>>> Mike
>>>
>>> -----Original Message-----
>>>       
>>>> From: Hector Centeno 
>>>> Sent: Oct 14, 2008 3:34 PM
>>>> To: csound@lists.bath.ac.uk
>>>> Subject: [Csnd] looking for advice on warming up phase vocoder processed sounds
>>>>
>>>> Hello,
>>>>
>>>> I've been using a lot the PVs family of opcodes to process sound
>>>> samples (doing time stretches, pitch shifting, smoothing, etc.) and
>>>> very often struggled trying to avoid "metallic" sounding results,
>>>> specially when doing time stretches (i.e. reading a PVX file with a
>>>> very slow pointer). One way I've been dealing with this is by using EQ
>>>> and reverb (tried also pvsmooth and pvsblur and didn't help much), but
>>>> I was wondering, are there any other techniques to avoid this? Is this
>>>> the result of analyzing using wrong parameters (I usually do: pvanal
>>>> -n 2048)? Do the PVS opcodes interpolate between windows? I've read in
>>>> some online articles about FFT in MAX/MSP and found mention to
>>>> interpolation as a way of increasing sound quality. Would this be a
>>>> way of avoiding it?
>>>>
>>>> Thanks!
>>>>
>>>> Hector
>>>>
>>>>
>>>> Send bugs reports to this list.
>>>> To unsubscribe, send email sympa@lists.bath.ac.uk with body "unsubscribe csound"
>>>>         
>>>
>>>
>>>
>>> Send bugs reports to this list.
>>> To unsubscribe, send email sympa@lists.bath.ac.uk with body "unsubscribe csound"
>>>
>>>       
>
>
> Send bugs reports to this list.
> To unsubscribe, send email sympa@lists.bath.ac.uk with body "unsubscribe csound"
>
>