[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Real-time pitch tracking

Dear John,

without having thought very carefully about this, I believe that the
definition of a real-time system does not depend on the latency (i.e.
the delay between input and corresponding output) it introduces.
For me, a real time system has to have two characteristics:
a) the system is causal
b) the required processing time per each input frame is lower or equals
the time span represented by that frame for all possible frame lengths

The latency, on the other hand, is just a property of a real time
system. It is extremely application-dependent what latency is tolerable,
so I don't believe you can simply define a threshold and say that one
system is real time because its latency is just below a threshold and
the next system is not because its latency is slightly above a threshold.

With respect to pitch tracking systems, I simply don't see how such
systems would theoretically be possible to design without latency. That
would mean you are able to extract the fundamental frequency from just
one sample of audio (and even then you would have 1 sample delay minimum
in reality but I would agree to refer to this as zero latency). Your
definition of zero latency, i.e. "locked in time with the onset of a
signal waveform, without delay" seems to be rather fuzzy to me.

Of course in real world computer systems, you will receive a block of
samples from the sound card; if you want to calculate just one result
for each block of samples, it doesn't matter at all whether your system
has a latency of block size or zero. But then you could also calculate a
DFT from this block of samples without adding latency.

Kind regards,

John Bates wrote:
> Dear List,
> Here's another YouTube pitch on pitch tracking.This time, Chris
> specializes in showing normal speech pitch tracking while describing
> some general information on the processing method.
>  http://www.youtube.com/watch?v=g2tohwbtFNA
> After the first Passaggio demonstration appeared, I received more
> information on other "real-time" pitch tracking methods. I see a problem
> here in the definition of real-time pitch display. In my lexicon,
> real-time processing means being locked in time with the onset of a
> signal waveform, without delay. The Passaggio pitch processor satisfies
> this criterion. On the other hand, I see no way that spectral processing
> can do this because of its windowing lag. This lag is not necessarily
> bad for the pitch display but it is a demonstration of the inherent
> problem in spectral methods for separating sound sources.  
> According  to Chris, there is some unnoticeable delay in Passaggio due
> to the need for "double buffering" to maintain compatibility with the
> Windows operating system. This delay is not inherent in the
> WIV-based processing method.
> John Bates