[PD] weird behavior of [vd~] in phave vocoder (overlapping subpatches)

[Alexandre Torres Porres]

I had said

"*So I was thinking and then thought that the other patch was working
because it wasn't a phase vocoder, so the phase alignment was not
important. But now that you say this and showed your patch, I can see that
you need not worry if you are using a delay. And I was actually using a
delay in my non fft patch.*

*In the end my patches were different but equivalent and now I get why.
It's cool to know and learn that if you are using delay you don't need to
care about it.*"

And I was just wrong! I wasn't using a delay line in the same way you were.
I just confused.

And the more I dig, the more my head hurts and the more confused I am... I
guess I'm back to square one...

Or worse, I guess I have more doubts now than at first :)

My first surprise was to see that if you had a delread~ in a parent patch
and a [vd~] into a subpatch with overlap is that it wouldn't generate
discontinuities... and I'm not sure why is that...

Now, you say

"*Having [delwrite~] and [vd~] in the same overlapping subpatch (as you
would in a spectral delay) is also not a problem. But having the
[delwrite~] in the overlapping subpatch and the [vd~] outside will cause
weirdness :-).*"

And I tested it. And hmm, I'm not sure what you mean, cause it only works
when you have a delread~ in a parent patch and a [vd~] into a subpatch with
overlap. I do have spectral delay patches and they just work, but if you
are listening to what comes out of both delread~ and vd~ in a subpatch,
it's just bad.

Check my attached patch. I don't really get why it works in the first one
and it doesn't in the other two. Maybe this is a first step before
venturing into the other implications of all this mess ;)

cheers



2015-09-10 22:53 GMT-03:00 Christof Ressi <christof.ressi at gmx.at>:

> Thanks for testing! I was suspecting that the difference might only be a
> very subtle one. But I'll check as well in next days. BTW: Your 'speed'
> control looks very cool, I'm gonna try this myself.
>
> I think I understand your questions better now, so I'll try to give some
> more concrete answers again:
>
> > My point was just that [vd~] acts in a special way when in an
> overlapping subpatch, and that is it'll output the audio
> > without discontinuities or pitch shifting because of interpreting the
> overlap as oversampling. That behaviour is special
> > when compared to [osc~], [phasor~]
>
> I don't understand what you mean here. [osc~] and [phasor~] also interpret
> the overlap as oversampling, as do all objects which rely on time
> information (ms, hz). In fact, overlapping is achieved by oversampling. The
> reason why there won't be any discontinuities with [vd~] is because it is
> only a reading object like [tabread4~] and the delay line itself is not
> affected by the overlapping. You only have to be careful when dealing with
> milliseconds and different sample rates. Having [delwrite~] and [vd~] in
> the same overlapping subpatch (as you would in a spectral delay) is also
> not a problem. But having the [delwrite~] in the overlapping subpatch and
> the [vd~] outside will cause weirdness :-).
>
> There are actual two 'problems' with [phasor~], [osc~] and [vline~] in
> overlapping subpatches:
>
> 1) looking from the outside they seem to run too slowly because they rely
> on a higher sample rate within in the subpatch, but contrary to deliberate
> upsampling, e. g. [block~ 64 1 4], the output doesn't get downsampled at
> the outlets. So with overlap 4 the sample rate is 176400 Hz instead of
> 44100 Hz. That means a [phasor~] with a speed of 44100 Hz has a period of 4
> samples. When it goes through the outlets it still has a period of 4
> samples but now the sample rate is 44100 Hz and its 'speed' is therefore
> interpreted as only 11025 Hz. You also have to be careful with milliseconds
> because they also depend on the sample rate.
> (Oddly enough, [samplerate~] always outputs the global samplerate and not
> the actual rate the subpatch is running at. This is why there is the
> [iem_samplerate~] object in iemlib, which always gives the actual
> samplerate.)
>
> 2) they run continously across blocks but because of overlapping they are
> not phase aligned after the outlet.
>
> The oversampling is the only reason for all the corrections you had to do
> in you patch. I attached a copy where I made some comments. I hope this
> helps. If you have any more questions you can ask me.
>
> Cheers
>
>
>
>
>
>
>
>
>
> *Gesendet:* Donnerstag, 10. September 2015 um 23:00 Uhr
> *Von:* "Alexandre Torres Porres" <porres at gmail.com>
> *An:* "Christof Ressi" <christof.ressi at gmx.at>
> *Cc:* Pd-List <pd-list at lists.iem.at>
> *Betreff:* Re: Re: Re: [PD] weird behavior of [vd~] in phave vocoder
> (overlapping subpatches)
> naaah, yeah, they're different.. oops... but doesn't really make any
> difference perceptually... let me check it some more...
>
> 2015-09-10 17:49 GMT-03:00 Alexandre Torres Porres <porres at gmail.com>:
>>
>> yeah, I have to sit again with some time and figure it out, I should do
>> some tests to better understand how many objects behave. But, in the
>> meantime, lets talk about something important here.
>>
>> > Delaying the back window [z~] is a rather lazy trick and it won't
>> > give accurate results each time you change the pitch shifting
>> > factor, but after one fft-window it settles. The question is if you
>> > can actually here this error. When I find some time I'll make a
>> > comparison between our both solutions.
>>
>> Are you really sure about this? Cause I've been testing it and thinking
>> about it and, in my opinion, both are exactly the same thing, equally
>> equivalent, and I can't hear any difference as well.
>>
>> Lets sort this out ;)
>>
>> I think that the second delay makes it a simpler patch and easier to
>> understand. I'm using [cyclone/delay~] by the way, which works with samples
>> - must be the same thing as [z~].
>>
>> cheers
>>
>> 2015-09-10 14:23 GMT-03:00 Christof Ressi <christof.ressi at gmx.at>:
>>>
>>> Hmmm, since we basically agree on all these things I was thinking if I
>>> missed a point, because I simply don't believe that [vd~] behaves
>>> differently than [tabread4~] and there is any unlogical or 'special'
>>> behaviour with [vd~] within an upsampled subpatch. Maybe one thing: The
>>> input of [vd~] is a time in milliseconds which is interpreted according to
>>> the actual sample rate (because internally the delay lines work on samples,
>>> of course). In that way it behaves like [phasor~], [vline~], [osc~]. So
>>> when you send 1000 ms to a [vd~] in a subpatch with overlap 4, the delay
>>> line will be read at sample 176400 (supposing the [delwrite~] is in a
>>> subpatch with sample rate 44100). This is not an issue if both the
>>> [delwrite~] and [vd~] are in the same subpatch. But if they are in
>>> subpatches with different sample rates you have to make some adjustments.
>>> If you're also aware of this behaviour than I don't know what else we
>>> could've missed...
>>>
>>> Cheers
>>> *Gesendet:* Donnerstag, 10. September 2015 um 18:10 Uhr
>>> *Von:* "Alexandre Torres Porres" <porres at gmail.com>
>>> *An:* "Christof Ressi" <christof.ressi at gmx.at>
>>> *Cc:* Pd-List <pd-list at lists.iem.at>
>>> *Betreff:* Re: Re: [PD] weird behavior of [vd~] in phave vocoder
>>> (overlapping subpatches)
>>> yeah, it'll consider the signal input is 0 so it'll output the
>>> corresponding index - which is "1" because of the interpolation.
>>>
>>> and yeah, I'm aware they're both buffer readers, delwrite~ / vd~ being a
>>> circular / ring buffer. And my point was this difference between them,
>>> where delay lines will always read/output at regular speed.
>>>
>>> But that is not the core of the discussion, and we actually agree on it,
>>> so I'm not sure what we're talking about here.
>>>
>>> My point was just that [vd~] acts in a special way when in an
>>> overlapping subpatch, and that is it'll output the audio without
>>> discontinuities or pitch shifting because of interpreting the overlap as
>>> oversampling. That behaviour is special when compared to [osc~], [phasor~]
>>> and I also tried a buffer reader like [tabplay~] and got "bad" results.
>>> They all don't work well in it, and so does not [vline~] by the way. There
>>> might be other relevant objects to test but I'm just not thinking about it.
>>> Nevertheless, I have the idea most will have problems, while some, like
>>> [vd~], will be be fine.
>>>
>>> The thing about [tabread~] is that it solely depends on external sources
>>> to read the buffers, while [vd~] doesn't, and that makes quite a practical
>>> difference in my opinion. The deal with [tabread~] is that the issue is
>>> more about what object is driving it and how it behaves (such as [vline~]
>>> and [phasor~], which don't behave well with overlapping subpatches).
>>>
>>> But again, not a relevant discussion. But I do feel like making more
>>> tests, I just don't know if there is a possible to test to check how the
>>> behaviour or [vd~] and [tabread4~] could relate between themselves.
>>>
>>> > For me its sometimes a trial and error game to find all
>>> > those parameters which have to be divided/multiplied
>>> > by the overlap factor. But after a while of thinking
>>> > everything turns out to make sense.
>>>
>>> yeah, it was trial and error, but I'm still not 100% sure how it makes
>>> sense... hence this thread :) - but I guess I'll keep thinking more about
>>> it.
>>>
>>> > Delaying the back window [z~] is a rather lazy trick and
>>> > it won't give accurate results each time you change the
>>> > pitch shifting factor,
>>>
>>> that's important to note, and that's why miller's patch may not have
>>> been using this procedure.
>>>
>>> thanks
>>>
>>>
>>> 2015-09-10 6:39 GMT-03:00 Christof Ressi <christof.ressi at gmx.at>:
>>>>
>>>> "Oops, now I'm the one to say you can't compare them as equal. You
>>>> see, [tabread~] needs to have an audio input to read through the array, but
>>>> [vd~] is always "reading" the buffer at normal speed - so this makes them
>>>> quite different. Since [tabread~] only react if some signal is feeding it,
>>>> then it depends solely on that incoming signal; and thus the object who's
>>>> outputting it, which could be [phasor~] or [vline~]. So it's more about
>>>> which object who's driving it than itself."
>>>>
>>>> Again, I insist that the behaviour of [tabread4~] and [vd~] is
>>>> equivalent ;-). When you don't feed any input to [tabread4~] it outputs the
>>>> value at index 1. Now try to think of a delay line as simply a table which
>>>> content is constantly updated at a time interval of 1/SR (SR = the actual
>>>> sample rate of the subpatch containing the [delwirte~]). If you don't send
>>>> any signal to [vd~], it behaves just as [tabread4~], only that the value at
>>>> index 1 always changes, so it only appears that [vd~] itself is reading
>>>> along a buffer. (Note that both objects can't read index 0 because of the
>>>> 4-point interpolation algorithm. So with [vd~] you will never get less than
>>>> a one sample delay.)
>>>> To make sloppy analogy:  [tabread4~] would be a band machine where the
>>>> tape itself stands still why the tape head can be freely moved, whereas
>>>> [vd~] would be one where the tape runs at a fixed speed and additionally
>>>> the tape head can be moved too. Well, I don't know if this makes sense :-).
>>>>
>>>> Since you took the word "reading" in quotation marks you might be aware
>>>> of all this. In that case the confusion might arise from the fact that you
>>>> have to consider the relation between the 'speed' of the delay line
>>>> (depending on the sample rate of the subpatch containing the [delwrite~])
>>>> and the 'speed' of the object providing the input for the [vd~].
>>>>
>>>> Please anyone correct me if I'm wrong on these points!
>>>>
>>>> For me its sometimes a trial and error game to find all those
>>>> parameters which have to be divided/multiplied by the overlap factor. But
>>>> after a while of thinking everything turns out to make sense.
>>>> Delaying the back window [z~] is a rather lazy trick and it won't give
>>>> accurate results each time you change the pitch shifting factor, but after
>>>> one fft-window it settles. The question is if you can actually here this
>>>> error. When I find some time I'll make a comparison between our both
>>>> solutions.
>>>>
>>>> Cheers, Christof
>>>>
>>>>
>>>>
>>>>
>>>> *Gesendet:* Donnerstag, 10. September 2015 um 07:51 Uhr
>>>> *Von:* "Alexandre Torres Porres" <porres at gmail.com>
>>>> *An:* "Christof Ressi" <christof.ressi at gmx.at>
>>>> *Cc:* Pd-List <pd-list at lists.iem.at>, "Gerd Schuller" <
>>>> studio at gerdschuller.com>
>>>> *Betreff:* Re: [PD] weird behavior of [vd~] in phave vocoder
>>>> (overlapping subpatches)
>>>> >>> unlike [osc~], [vd~] will get the continuities between blocks
>>>> right!"
>>>>
>>>> > You can't really compare these two objects.
>>>>
>>>> Sure I can :) i'll insist on it by the way. Again, [vd~] will not
>>>> generate discontinuities with the overlaps, unlike other objects such as
>>>> [osc~] and [phasor~]. Moreover, and as a logical result, it won't change
>>>> the pitch because of the oversampling. It'll just work fine.
>>>>
>>>> > [vd~] is actually the same thing as [tabread4~]
>>>>
>>>> Oops, now I'm the one to say you can't compare them as equal. You see,
>>>> [tabread~] needs to have an audio input to read through the array, but
>>>> [vd~] is always "reading" the buffer at normal speed - so this makes them
>>>> quite different. Since [tabread~] only react if some signal is feeding it,
>>>> then it depends solely on that incoming signal; and thus the object who's
>>>> outputting it, which could be [phasor~] or [vline~]. So it's more about
>>>> which object who's driving it than itself.
>>>>
>>>> When it comes to [vd~], the pithc shifting and time stretching also
>>>> depends on the object that's driving the input, which could be again
>>>> [phasor~] or [vline~] and need to deal with their behaviour.
>>>>
>>>> > you have to divide by the overlap factor, because then
>>>> > you read less samples and therefore virtually slow the
>>>> > [vline~] down. In a subpatch with overlap 4 everything
>>>> > happens 4 times as fast because instead of only 1 block
>>>>
>>>> yeah, sure, I've pointed it in my 1st message. I get that.
>>>>
>>>> But as I asked, I don't really get how ALL parameters need to divided
>>>> by 4, not only the [vline~] time, that is not clear yet. Sorry.
>>>>
>>>> And by the way, my patch does also time stretching, so it's different
>>>> than yours and is dealing with more parameters and issues than you. So you
>>>> are addressing the [vline~] issue only (replaced by [phasor~] in your
>>>> patch) - but that was the only parameter that I really understood anyway.
>>>>
>>>> > When using [z~] to delay the back window simply by the
>>>> > fft hop size you don't have to care about  window sizes
>>>>
>>>> hmm, my problem was more why were my two patches different, the one
>>>> with fft needed to care about it, but the other one didn't. I actually get
>>>> why that thing needs to be done to properly phase align the windows.
>>>>
>>>> So I was thinking and then thought that the other patch was working
>>>> because it wasn't a phase vocoder, so the phase alignment was not
>>>> important. But now that you say this and showed your patch, I can see that
>>>> you need not worry if you are using a delay. And I was actually using a
>>>> delay in my non fft patch.
>>>>
>>>> In the end my patches were different but equivalent and now I get why.
>>>> It's cool to know and learn that if you are using delay you don't need to
>>>> care about it.
>>>>
>>>> thanks
>>>>
>>>> ps. I'm still curious on sorting out the behaviour of [vd~] though
>>>>
>>>>
>>>>
>>>> 2015-09-09 7:54 GMT-03:00 Christof Ressi <christof.ressi at gmx.at>:
>>>>>
>>>>> Hi Alexandre,
>>>>>
>>>>> I'm new on this list, but I think I can help you on this because
>>>>> recently I tried to do the same thing. I can't fully test your patch
>>>>> because I'm missing the cyclone library (and don't bother to install it
>>>>> :-p).  I try to give an answer to the following questions:
>>>>>
>>>>> "Other issues related to overlapping besides this "oversampling" is
>>>>> that some objects won't make it right, they'll chop the blocks with
>>>>> discontinuities, such as the case with [osc~]. But as it turns out, unlike
>>>>> [osc~], [vd~] will get the continuities between blocks right!"
>>>>>
>>>>> You can't really compare these two objects. [vd~] is actually the same
>>>>> thing as [tabread4~], only that it reads from a ring buffer rather from a
>>>>> table. So the critical thing is only which object you use as the input for
>>>>> [vd~]. You are using [vline~] whereas I'm using [phasor~]. Both are
>>>>> equivalent. For the reading index for [vd~] you have to divide by the
>>>>> overlap factor, because then you read less samples and therefore virtually
>>>>> slow the [vline~] down. In a subpatch with overlap 4 everything happens 4
>>>>> times as fast because instead of only 1 block, 4 blocks have to be
>>>>> processed - in the same time!). My approach is to have a [phasor~] run from
>>>>> 0 to 1 (or 1 to 0) for every block so I have to multiply it's speed by
>>>>> four. Than I multiply the output by the windows size. Note that in my patch
>>>>> I get the second window one hop size behind by simply delaying it with [z~]
>>>>> whereas you've chosen to use a second [vd~] with a wrapping object. (I
>>>>> guess you're way actually saves some memory as you don't need a second
>>>>> delay line).
>>>>>
>>>>> "And even more weirdly, in the Pvoc patch I have to multiply the
>>>>> difference between the front and back windows to the ratio of
>>>>> transposition. This is even crazier than the last issue, and I have no idea
>>>>> why that has to be this way..."
>>>>>
>>>>> When you're transposing you're actually reading more samples for
>>>>> upwards pitchshifting and less samples for downwards pitchshifting. So you
>>>>> basically stretch or compress the window size. This means also that the
>>>>> time difference between two windows changes if you want them to be phase
>>>>> aligned. If the window gets larger, the time difference to the last window
>>>>> also gets larger and vice verca. You might be aware of this: The window in
>>>>> the back has to be phase aligned with the front window because you need it
>>>>> as a reference to calculate the difference from the actual phase of the
>>>>> previous output window.
>>>>>
>>>>> When using [z~] to delay the back window simply by the fft hop size
>>>>> you don't have to care about window sizes and time differences at all. It
>>>>> is, however, also a bit incorrect for the first analysis window after a
>>>>> change of pitch so I might change it and try it your way!
>>>>>
>>>>> You can have a look at my solution and compare it to yours. From what
>>>>> I've seen both work the same way though I couldn't test your patch.
>>>>> However, I think that my patch could be conceptually easier to understand,
>>>>> but I might be wrong :-).
>>>>>
>>>>> Cheers, Christof
>>>>>
>>>>> PS: Ignore the right half of [pd read-windows] with the two
>>>>> [tabread4~], this is only needed for the freeze effect.
>>>>>
>>>>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.puredata.info/pipermail/pd-list/attachments/20150911/8864bd20/attachment-0001.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: good-bad_delay-tests.pd
Type: application/octet-stream
Size: 1031 bytes
Desc: not available
URL: <http://lists.puredata.info/pipermail/pd-list/attachments/20150911/8864bd20/attachment-0001.obj>