Re: [PD] Non-Linear Quantization / Bitcrush
On Tue, Nov 02, 2010 at 06:13:49PM -0400, brandon zeeb wrote: Great! Unfortunately, I'm not entirely sure how to realize this in Pd. Can you help me out with a little example? Like in the attachement maybe. The [pd roundit] doesn't do a floor, but a real rounding. Rounding or flooring is at the core of bitcrusher effects as in the rj library, too, so you should have no problem to apply this to signals. Ciao -- Frank BarknechtDo You RjDj.me? _ __footils.org__ quantized.pd Description: application/puredata ___ Pd-list@iem.at mailing list UNSUBSCRIBE and account-management - http://lists.puredata.info/listinfo/pd-list
Re: [PD] Non-Linear Quantization / Bitcrush
could you give examples of idealized input and output for cases 1-4? im not sure I understand what exactly you want... interested greetings! Ludwig On 1 November 2010 13:09, brandon zeeb zeeb.bran...@gmail.com wrote: Hey All, I've been burning my brain over this issue lately and I can't seem to come up with an elegant solution, and stay with me here as I attempt to explain it best I can. For me and my needs, being able to quantize an arbitrary signal to any arbitrary series is the Holy Grail (and I'm not talking about simple table lookup!). I'm looking to quantize an incoming signal (or value) given a max and min quantization value and an arbitrary curve. Think quantization of note events to a series of note lengths or your standard bitcrush algorithm, it's pretty much the same. The arbitrary curve should influence the degree to which the bitcrush algorithm is applied to the signal such that one could have less quantization at smaller values of the input signal, and greater quantization and larger values (or vice versa). Simple table-lookup is insufficient as it requires you to pre-define a maximum input signal amount. I'm willing to waive this requirement if an implementation is not possible without it. This will be used in the following circumstances: To quantize envelopes signals to any arbitrary series (say !, Fibonacci, x^2, 2^x, etc) To quantize signal loop length values to an arbitrary series of note values (say 1/16, 1/8, 1/2, 1/1) To apply non-linear bitcrushing to a signal such that higher values are expressed with less of an effect than smaller values To quantize pitch events to a pre-defined series Is this making sense? My attempts thus far has extended the RjDj bitcrush abstraction with mild success. I can recreate the effect but the output signal bears too many artifacts from the input signal (ie: the curve retains part of it's original slope from the input signal and is not flattened or held until the next value). Thanks, ~Brandon ___ Pd-list@iem.at mailing list UNSUBSCRIBE and account-management - http://lists.puredata.info/listinfo/pd-list ___ Pd-list@iem.at mailing list UNSUBSCRIBE and account-management - http://lists.puredata.info/listinfo/pd-list
Re: [PD] Non-Linear Quantization / Bitcrush
I've attached my best attempt at recreating this effect, the attached PNG will be used as a reference. Given the distance d1 and d2, these distances are usually identical in a traditional bitcrush or simple quantization. I would like to be able to vary the distance between points of an incoming signal such that the distance between points is a function of a given quantization value, the current Y value, AND a given quantization curve. As a first step (and illustrated in the attached PNG), smaller values of Y will produce a more pronounced quantization given a x^2 quantization curve while larger values of Y will produce a smaller distance between steps. Ideas? ~Brandon On Tue, Nov 2, 2010 at 10:47 AM, Ludwig Maes ludwig.m...@gmail.com wrote: could you give examples of idealized input and output for cases 1-4? im not sure I understand what exactly you want... interested greetings! Ludwig On 1 November 2010 13:09, brandon zeeb zeeb.bran...@gmail.com wrote: Hey All, I've been burning my brain over this issue lately and I can't seem to come up with an elegant solution, and stay with me here as I attempt to explain it best I can. For me and my needs, being able to quantize an arbitrary signal to any arbitrary series is the Holy Grail (and I'm not talking about simple table lookup!). I'm looking to quantize an incoming signal (or value) given a max and min quantization value and an arbitrary curve. Think quantization of note events to a series of note lengths or your standard bitcrush algorithm, it's pretty much the same. The arbitrary curve should influence the degree to which the bitcrush algorithm is applied to the signal such that one could have less quantization at smaller values of the input signal, and greater quantization and larger values (or vice versa). Simple table-lookup is insufficient as it requires you to pre-define a maximum input signal amount. I'm willing to waive this requirement if an implementation is not possible without it. This will be used in the following circumstances: To quantize envelopes signals to any arbitrary series (say !, Fibonacci, x^2, 2^x, etc) To quantize signal loop length values to an arbitrary series of note values (say 1/16, 1/8, 1/2, 1/1) To apply non-linear bitcrushing to a signal such that higher values are expressed with less of an effect than smaller values To quantize pitch events to a pre-defined series Is this making sense? My attempts thus far has extended the RjDj bitcrush abstraction with mild success. I can recreate the effect but the output signal bears too many artifacts from the input signal (ie: the curve retains part of it's original slope from the input signal and is not flattened or held until the next value). Thanks, ~Brandon ___ Pd-list@iem.at mailing list UNSUBSCRIBE and account-management - http://lists.puredata.info/listinfo/pd-list -- Brandon Zeeb Columbus, Ohio attachment: my-crush-ex.png my-crush.pd Description: Binary data ___ Pd-list@iem.at mailing list UNSUBSCRIBE and account-management - http://lists.puredata.info/listinfo/pd-list
Re: [PD] Non-Linear Quantization / Bitcrush
This is even better. If I could minimize the jumps around Y = 0.5 to -0.5 It'll be exactly what I'm looking for... or a start at least. Do you see what I mean now? See how the amount of quantization changes with Y and a minimum quantization value? I think I'm getting towards the answer now... -- Brandon Zeeb Columbus, Ohio attachment: my-crush2-ex.png my-crush2.pd Description: Binary data ___ Pd-list@iem.at mailing list UNSUBSCRIBE and account-management - http://lists.puredata.info/listinfo/pd-list
Re: [PD] Non-Linear Quantization / Bitcrush
So you want amplitude 'a' dependant quantization size 'q' ? take your chosen q(a); in your example it seems you want a simple line: q=q(0)-k*a; define f(a) as integral of 1/q from a=0 to a; also calculate the inverse of f(a) i.e. a(f); now for each sample do: out=a(round(f(in))) where round is any floor or the like... have fun! ps: in your example: q=q0-k*a with for example q(0)=0.001 and q(0.8)=0.0001: q:=0.001-0.0009/0.8*a then f=2558.427881-.11*ln(10.-9.*a) and inverse=easy On 2 November 2010 19:20, Ludwig Maes ludwig.m...@gmail.com wrote: This is pretty easy actually, I use such things mostly to guide my rhythmical quantization... On 2 November 2010 19:19, brandon zeeb zeeb.bran...@gmail.com wrote: This is even better. If I could minimize the jumps around Y = 0.5 to -0.5 It'll be exactly what I'm looking for... or a start at least. Do you see what I mean now? See how the amount of quantization changes with Y and a minimum quantization value? I think I'm getting towards the answer now... -- Brandon Zeeb Columbus, Ohio ___ Pd-list@iem.at mailing list UNSUBSCRIBE and account-management - http://lists.puredata.info/listinfo/pd-list
Re: [PD] Non-Linear Quantization / Bitcrush
The reason you use the inverse is so that the amplitude remains the same albeit quantized. The reason we use another function before flooring is to distritube the floor levels.But afterwards we need to bring the values back to their original place On 2 November 2010 19:37, Ludwig Maes ludwig.m...@gmail.com wrote: So you want amplitude 'a' dependant quantization size 'q' ? take your chosen q(a); in your example it seems you want a simple line: q=q(0)-k*a; define f(a) as integral of 1/q from a=0 to a; also calculate the inverse of f(a) i.e. a(f); now for each sample do: out=a(round(f(in))) where round is any floor or the like... have fun! ps: in your example: q=q0-k*a with for example q(0)=0.001 and q(0.8)=0.0001: q:=0.001-0.0009/0.8*a then f=2558.427881-.11*ln(10.-9.*a) and inverse=easy On 2 November 2010 19:20, Ludwig Maes ludwig.m...@gmail.com wrote: This is pretty easy actually, I use such things mostly to guide my rhythmical quantization... On 2 November 2010 19:19, brandon zeeb zeeb.bran...@gmail.com wrote: This is even better. If I could minimize the jumps around Y = 0.5 to -0.5 It'll be exactly what I'm looking for... or a start at least. Do you see what I mean now? See how the amount of quantization changes with Y and a minimum quantization value? I think I'm getting towards the answer now... -- Brandon Zeeb Columbus, Ohio ___ Pd-list@iem.at mailing list UNSUBSCRIBE and account-management - http://lists.puredata.info/listinfo/pd-list
Re: [PD] Non-Linear Quantization / Bitcrush
And we want f' to be 1 (integer step) / (per) quantization size (for that amplitude) On 2 November 2010 19:41, Ludwig Maes ludwig.m...@gmail.com wrote: The reason you use the inverse is so that the amplitude remains the same albeit quantized. The reason we use another function before flooring is to distritube the floor levels.But afterwards we need to bring the values back to their original place On 2 November 2010 19:37, Ludwig Maes ludwig.m...@gmail.com wrote: So you want amplitude 'a' dependant quantization size 'q' ? take your chosen q(a); in your example it seems you want a simple line: q=q(0)-k*a; define f(a) as integral of 1/q from a=0 to a; also calculate the inverse of f(a) i.e. a(f); now for each sample do: out=a(round(f(in))) where round is any floor or the like... have fun! ps: in your example: q=q0-k*a with for example q(0)=0.001 and q(0.8)=0.0001: q:=0.001-0.0009/0.8*a then f=2558.427881-.11*ln(10.-9.*a) and inverse=easy On 2 November 2010 19:20, Ludwig Maes ludwig.m...@gmail.com wrote: This is pretty easy actually, I use such things mostly to guide my rhythmical quantization... On 2 November 2010 19:19, brandon zeeb zeeb.bran...@gmail.com wrote: This is even better. If I could minimize the jumps around Y = 0.5 to -0.5 It'll be exactly what I'm looking for... or a start at least. Do you see what I mean now? See how the amount of quantization changes with Y and a minimum quantization value? I think I'm getting towards the answer now... -- Brandon Zeeb Columbus, Ohio ___ Pd-list@iem.at mailing list UNSUBSCRIBE and account-management - http://lists.puredata.info/listinfo/pd-list
Re: [PD] Non-Linear Quantization / Bitcrush
Watch out in my numeric example, I was a bit careless and the q I chose continues to increase for more and more negative amplitudes! On 2 November 2010 19:44, Ludwig Maes ludwig.m...@gmail.com wrote: And we want f' to be 1 (integer step) / (per) quantization size (for that amplitude) On 2 November 2010 19:41, Ludwig Maes ludwig.m...@gmail.com wrote: The reason you use the inverse is so that the amplitude remains the same albeit quantized. The reason we use another function before flooring is to distritube the floor levels.But afterwards we need to bring the values back to their original place On 2 November 2010 19:37, Ludwig Maes ludwig.m...@gmail.com wrote: So you want amplitude 'a' dependant quantization size 'q' ? take your chosen q(a); in your example it seems you want a simple line: q=q(0)-k*a; define f(a) as integral of 1/q from a=0 to a; also calculate the inverse of f(a) i.e. a(f); now for each sample do: out=a(round(f(in))) where round is any floor or the like... have fun! ps: in your example: q=q0-k*a with for example q(0)=0.001 and q(0.8)=0.0001: q:=0.001-0.0009/0.8*a then f=2558.427881-.11*ln(10.-9.*a) and inverse=easy On 2 November 2010 19:20, Ludwig Maes ludwig.m...@gmail.com wrote: This is pretty easy actually, I use such things mostly to guide my rhythmical quantization... On 2 November 2010 19:19, brandon zeeb zeeb.bran...@gmail.com wrote: This is even better. If I could minimize the jumps around Y = 0.5 to -0.5 It'll be exactly what I'm looking for... or a start at least. Do you see what I mean now? See how the amount of quantization changes with Y and a minimum quantization value? I think I'm getting towards the answer now... -- Brandon Zeeb Columbus, Ohio ___ Pd-list@iem.at mailing list UNSUBSCRIBE and account-management - http://lists.puredata.info/listinfo/pd-list
Re: [PD] Non-Linear Quantization / Bitcrush
Great! Unfortunately, I'm not entirely sure how to realize this in Pd. Can you help me out with a little example? Thanks On Tue, Nov 2, 2010 at 2:37 PM, Ludwig Maes ludwig.m...@gmail.com wrote: So you want amplitude 'a' dependant quantization size 'q' ? take your chosen q(a); in your example it seems you want a simple line: q=q(0)-k*a; define f(a) as integral of 1/q from a=0 to a; also calculate the inverse of f(a) i.e. a(f); now for each sample do: out=a(round(f(in))) where round is any floor or the like... have fun! ps: in your example: q=q0-k*a with for example q(0)=0.001 and q(0.8)=0.0001: q:=0.001-0.0009/0.8*a then f=2558.427881-.11*ln(10.-9.*a) and inverse=easy On 2 November 2010 19:20, Ludwig Maes ludwig.m...@gmail.com wrote: This is pretty easy actually, I use such things mostly to guide my rhythmical quantization... On 2 November 2010 19:19, brandon zeeb zeeb.bran...@gmail.com wrote: This is even better. If I could minimize the jumps around Y = 0.5 to -0.5 It'll be exactly what I'm looking for... or a start at least. Do you see what I mean now? See how the amount of quantization changes with Y and a minimum quantization value? I think I'm getting towards the answer now... -- Brandon Zeeb Columbus, Ohio -- Brandon Zeeb Columbus, Ohio ___ Pd-list@iem.at mailing list UNSUBSCRIBE and account-management - http://lists.puredata.info/listinfo/pd-list
[PD] Non-Linear Quantization / Bitcrush
Hey All, I've been burning my brain over this issue lately and I can't seem to come up with an elegant solution, and stay with me here as I attempt to explain it best I can. For me and my needs, being able to quantize an arbitrary signal to any arbitrary series is the Holy Grail (and I'm not talking about simple table lookup!). I'm looking to quantize an incoming signal (or value) given a max and min quantization value and an arbitrary curve. Think quantization of note events to a series of note lengths or your standard bitcrush algorithm, it's pretty much the same. The arbitrary curve should influence the degree to which the bitcrush algorithm is applied to the signal such that one could have less quantization at smaller values of the input signal, and greater quantization and larger values (or vice versa). Simple table-lookup is insufficient as it requires you to pre-define a maximum input signal amount. I'm willing to waive this requirement if an implementation is not possible without it. This will be used in the following circumstances: 1. To quantize envelopes signals to any arbitrary series (say !, Fibonacci, x^2, 2^x, etc) 2. To quantize signal loop length values to an arbitrary series of note values (say 1/16, 1/8, 1/2, 1/1) 3. To apply non-linear bitcrushing to a signal such that higher values are expressed with less of an effect than smaller values 4. To quantize pitch events to a pre-defined series Is this making sense? My attempts thus far has extended the RjDj bitcrush abstraction with mild success. I can recreate the effect but the output signal bears too many artifacts from the input signal (ie: the curve retains part of it's original slope from the input signal and is not flattened or held until the next value). Thanks, ~Brandon ___ Pd-list@iem.at mailing list UNSUBSCRIBE and account-management - http://lists.puredata.info/listinfo/pd-list
