Hi, there. 2 years ago, I created this thread on the forum: http://forum.pdpatchrepo.info/topic/7340/echo-supressor. After talking about dBs, I thought I acchived what I was looking for. But no. Nor close.
My last non-replied post on september 2014 says:
/Bringing back to life an old thread... I tried to make this patch work LIVE... and it is not as thought./
/To make this patch work (*1), I should find the perfect delay to suppress the audio. Using 44100, the delay time can vary in 0.0226ms. So... it's almost impossible to get that. I only acchieve a flanger, nothing else./
/In the first Patch-Circle done in Argentina last Saturday, I talked with Pablo E. Riera about this and he told me about the adaptive external and the echo cancellation example... but there is no example like that (only intereference cancellation)./
/I will explain again (*2): the operator has a microphone (using one channel of input) and it sounds in a LCD (far away) (using one channel of output). Above the LCD I have a mic (second channel of input). This mic takes what the audience says and send to the operator headphones (second channel of output). The operator hears her own voice (with a delay). I want to supress this "return". It doesn't matter the latency, I just don't want the operator to hear her/him voice./
/I would need something that learn from the first microphone and get rid of what the second mic takes./
/I hope being clear in my explanation. Thanks in advace./
//
(*1): the patch is in the thread. It does something like this: it takes the microphone (channel one), invert the wave (by using [*~ -1]), it uses a delayline to synch (because of the latency) and mix with the output. The intention is, obviously, suppress the microphone after been listened by the second microphone.
[adc~]
|
| [delwrite~ echo-supressor-A 1000]
|
[delwrite~ echo-supressor-B 1000]
[Hslider] # to synch. It depends on latency (it would have to be exact, here is the problem!) | [vd~ echo_supressor-A] | [*~ -1] | | [vd~ echo_supressor-B 1] | / [+~] | [dac~]
(*2): I just updated the explanation
How can I acchieve an echo suppresor? Is this possible? If I'm not clear in my explanation, please tell me. Thanks.
- Mario Mey mariomey@gmail.com [2015-05-04 10:18]:
[...]
How can I acchieve an echo suppresor? Is this possible?
Good acoustic echo cancellation (AEC) is quite a substantial topic. Perhaps http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=1146062 could serve as a good introductory text. Search the usual scientific databases for the name of that field of study for more materiel.
You're not likely to get very good results with just a single delay. The way these systems normally work (e.g. on phones and conferencing systems) is to continuously estimate the transfer function between the speaker and the microphone (impulse response in the time domain). Then you take the received signal, convolve it with the inverse of the impulse response, and mix it into the signal you receive from the microphone.
The diagram at the bottom of this page[1] is a pretty good one to understand what's going on.
-s
http://www.speex.org/docs/manual/speex-manual/node4.html
On Mon, May 4, 2015, at 10:18 AM, Mario Mey wrote:
Hi, there. 2 years ago, I created this thread on the forum:
http://forum.pdpatchrepo.info/topic/7340/echo-supressor. After
talking about dBs, I thought I acchived what I was looking for. But
no. Nor close.
My last non-replied post on september 2014 says:
*Bringing back to life an old thread... I tried to make
this patch work LIVE... and it is not as thought.*
*To make this patch work (*1), I should find the perfect
delay to suppress the audio. Using 44100, the delay time can
vary in 0.0226ms. So... it's almost impossible to get that.
I only acchieve a flanger, nothing else.*
*In the first Patch-Circle done in Argentina last Saturday,
I talked with Pablo E. Riera about this and he told me about
the adaptive external and the echo cancellation example...
but there is no example like that (only intereference
cancellation).*
*I will explain again (*2): the operator has a microphone
(using one channel of input) and it sounds in a LCD (far
away) (using one channel of output). Above the LCD I have a
mic (second channel of input). This mic takes what the
audience says and send to the operator headphones (second
channel of output). The operator hears her own voice (with
a delay). I want to supress this "return". It doesn't
matter the latency, I just don't want the operator to hear
her/him voice.*
*I would need something that learn from the first
microphone and get rid of what the second mic takes.*
*I hope being clear in my explanation. Thanks in advace.*
(*1): the patch is in the thread. It does something like this: it
takes the microphone (channel one), invert the wave (by using [*~
-1]), it uses a delayline to synch (because of the latency) and mix
with the output. The intention is, obviously, suppress the
microphone after been listened by the second microphone.
[adc~]
| \
| [delwrite~ echo-supressor-A 1000]
|
[delwrite~ echo-supressor-B 1000]
[Hslider] # to synch. It depends on latency (it would have to be
exact, here is the problem!)
|
[vd~ echo_supressor-A]
|
[*~ -1]
|
|[vd~ echo_supressor-B 1]
|/
[+~]
|
[dac~]
(*2): I just updated the explanation
How can I acchieve an echo suppresor? Is this possible?
If I'm not clear in my explanation, please tell me. Thanks.
Pd-list@lists.iem.at mailing list UNSUBSCRIBE and account-management -> http://lists.puredata.info/listinfo/pd-list
Yes, as I said in the mail, not only the the time of the delay is very very short that it would be very difficult for me to find "at hand", also, as you say, this time would be variable. My system don't work... I knew it.
I took a look to some pages (*), but I only understand... the title of the mails and the PDF filename. The texts are chinese for me. I didn't know that echo cancellation is, as Peter Parker said, "quite a substantial topic"... and it is out of my hands to try different EC systems. No time, no knowledge, no time... no time.
Also, there's http://grh.mur.at/software/adaptive.html... but I don't know how to test it.
I made a diagram of my system... if anyone knows which echo cancellation system would work for me... I would reallly appreciate it.
Thank you.
(*) http://comments.gmane.org/gmane.comp.multimedia.puredata.general/63658 https://ccrma.stanford.edu/~eberdahl/Projects/FeedbackCancellation/FeedbackC... Peter Parker sent me this link, but I have to pay to see it: http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=1146062.
El 04/05/15 a las 13:48, Spencer Russell escibió:
You're not likely to get very good results with just a single delay. The way these systems normally work (e.g. on phones and conferencing systems) is to continuously estimate the transfer function between the speaker and the microphone (impulse response in the time domain). Then you take the received signal, convolve it with the inverse of the impulse response, and mix it into the signal you receive from the microphone.
The diagram at the bottom of this page[1] is a pretty good one to understand what's going on.
-s
http://www.speex.org/docs/manual/speex-manual/node4.html On Mon, May 4, 2015, at 10:18 AM, Mario Mey wrote:
Hi, there. 2 years ago, I created this thread on the forum: http://forum.pdpatchrepo.info/topic/7340/echo-supressor. After talking about dBs, I thought I acchived what I was looking for. But no. Nor close. My last non-replied post on september 2014 says:
/Bringing back to life an old thread... I tried to make this patch work LIVE... and it is not as thought./
/To make this patch work (*1), I should find the perfect delay to suppress the audio. Using 44100, the delay time can vary in 0.0226ms. So... it's almost impossible to get that. I only acchieve a flanger, nothing else./
/In the first Patch-Circle done in Argentina last Saturday, I talked with Pablo E. Riera about this and he told me about the adaptive external and the echo cancellation example... but there is no example like that (only intereference cancellation)./
/I will explain again (*2): the operator has a microphone (using one channel of input) and it sounds in a LCD (far away) (using one channel of output). Above the LCD I have a mic (second channel of input). This mic takes what the audience says and send to the operator headphones (second channel of output). The operator hears her own voice (with a delay). I want to supress this "return". It doesn't matter the latency, I just don't want the operator to hear her/him voice./
/I would need something that learn from the first microphone and get rid of what the second mic takes./
/I hope being clear in my explanation. Thanks in advace./
(*1): the patch is in the thread. It does something like this: it takes the microphone (channel one), invert the wave (by using [*~ -1]), it uses a delayline to synch (because of the latency) and mix with the output. The intention is, obviously, suppress the microphone after been listened by the second microphone.
[adc~] |
| [delwrite~ echo-supressor-A 1000] | [delwrite~ echo-supressor-B 1000][Hslider] # to synch. It depends on latency (it would have to be exact, here is the problem!) | [vd~ echo_supressor-A] | [*~ -1] | | [vd~ echo_supressor-B 1] | / [+~] | [dac~]
(*2): I just updated the explanation How can I acchieve an echo suppresor? Is this possible? If I'm not clear in my explanation, please tell me. Thanks. _________________________________________________ Pd-list@lists.iem.at mailto:Pd-list@lists.iem.at mailing list UNSUBSCRIBE and account-management -> http://lists.puredata.info/listinfo/pd-list
Pd-list@lists.iem.at mailing list UNSUBSCRIBE and account-management -> http://lists.puredata.info/listinfo/pd-list
Hi Mario,
I know how to solve this problem, but I'm only part-way there. I have a similar problem, rejecting feedback when processing acoustic instruments. I'm sorry I didn't post what I've got sooner--it needs some more work. I can't really tell if I'll be finished with a decent patch soon.
I'm almost through with the derivation in high-level symbolic math, so you can follow along with the patch and work on it too.
See the latency tester first. It plays noise from to [dac~] and pics up the signal from [adc~]. If you want to get a really clean graph for measuring your system latency, do a loopback test with the cable running from output to input on your audio interface. It should run reasonably well over loudspeakers and will tell you the round trip latency. There are two sliders where you can adjust delay to the input (mic) signal or the output (speaker) signal.
Next, there's the spectrum analyzer patch. Open the spect-2048_test.pd patch, then right click on [spect-2048] and open it. This solves the spectrum analysis problem on block size 2048. The result is just written to some tables in [spect-2048], for you to visually see it.
When it's set to run on block size 16384, it needs to have about 90-100 ms of latency (don't even try opening it, without the audio buffer length turned up).
Once you've captured an impulse with the spectrum analyzer as it is now, you'd have to save the impulse to a file and then re-open with your runtime patch (and lower latency settings). The patch spect-16384_test.pd patch does a better job of calculating the impulse response than the 2048 block size patch.
I'd like to have reduced latency to run the spectrum analyzer patch with larger block sizes (for which I need a different [xcov~] that works well on overlapped block sizes). The syntax of [xcov~] is already so ugly and will get worse with overlapped blocks.
Also, the spectrum analyzer math needs a little work. More to come...
Chuck
On May 7, 2015 7:15 AM, "Mario Mey" mariomey@gmail.com wrote:
Yes, as I said in the mail, not only the the time of the delay is very very short that it would be very difficult for me to find "at hand", also, as you say, this time would be variable. My system don't work... I knew it.
I took a look to some pages (*), but I only understand... the title of the mails and the PDF filename. The texts are chinese for me. I didn't know that echo cancellation is, as Peter Parker said, "quite a substantial topic"... and it is out of my hands to try different EC systems. No time, no knowledge, no time... no time.
Also, there's http://grh.mur.at/software/adaptive.html... but I don't know how to test it.
I made a diagram of my system... if anyone knows which echo cancellation system would work for me... I would reallly appreciate it.
Thank you.
(*) http://comments.gmane.org/gmane.comp.multimedia.puredata.general/63658
https://ccrma.stanford.edu/~eberdahl/Projects/FeedbackCancellation/FeedbackC... Peter Parker sent me this link, but I have to pay to see it: http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=1146062.
El 04/05/15 a las 13:48, Spencer Russell escibió:
You're not likely to get very good results with just a single delay. The way these systems normally work (e.g. on phones and conferencing systems) is to continuously estimate the transfer function between the speaker and the microphone (impulse response in the time domain). Then you take the received signal, convolve it with the inverse of the impulse response, and mix it into the signal you receive from the microphone.
The diagram at the bottom of this page[1] is a pretty good one to understand what's going on.
-s http://www.speex.org/docs/manual/speex-manual/node4.html
On Mon, May 4, 2015, at 10:18 AM, Mario Mey wrote:
Hi, there. 2 years ago, I created this thread on the forum: http://forum.pdpatchrepo.info/topic/7340/echo-supressor. After talking about dBs, I thought I acchived what I was looking for. But no. Nor close.
My last non-replied post on september 2014 says:
*Bringing back to life an old thread... I tried to make this patch work LIVE... and it is not as thought.*
*To make this patch work (*1), I should find the perfect delay to suppress the audio. Using 44100, the delay time can vary in 0.0226ms. So... it's almost impossible to get that. I only acchieve a flanger, nothing else.*
*In the first Patch-Circle done in Argentina last Saturday, I talked with Pablo E. Riera about this and he told me about the adaptive external and the echo cancellation example... but there is no example like that (only intereference cancellation).*
*I will explain again (*2): the operator has a microphone (using one channel of input) and it sounds in a LCD (far away) (using one channel of output). Above the LCD I have a mic (second channel of input). This mic takes what the audience says and send to the operator headphones (second channel of output). The operator hears her own voice (with a delay). I want to supress this "return". It doesn't matter the latency, I just don't want the operator to hear her/him voice.*
*I would need something that learn from the first microphone and get rid of what the second mic takes.*
*I hope being clear in my explanation. Thanks in advace.*
(*1): the patch is in the thread. It does something like this: it takes the microphone (channel one), invert the wave (by using [*~ -1]), it uses a delayline to synch (because of the latency) and mix with the output. The intention is, obviously, suppress the microphone after been listened by the second microphone.
[adc~] |
| [delwrite~ echo-supressor-A 1000] | [delwrite~ echo-supressor-B 1000][Hslider] # to synch. It depends on latency (it would have to be exact, here is the problem!) | [vd~ echo_supressor-A] | [*~ -1] | | [vd~ echo_supressor-B 1] | / [+~] | [dac~] (*2): I just updated the explanation
How can I acchieve an echo suppresor? Is this possible? If I'm not clear in my explanation, please tell me. Thanks. *_______________________________________________* Pd-list@lists.iem.at mailing list UNSUBSCRIBE and account-management -> http://lists.puredata.info/listinfo/pd-list
_______________________________________________Pd-list@lists.iem.at mailing list UNSUBSCRIBE and account-management -> http://lists.puredata.info/listinfo/pd-list
Pd-list@lists.iem.at mailing list UNSUBSCRIBE and account-management -> http://lists.puredata.info/listinfo/pd-list
Sorry--I left out this one: xcov-2048
On Thu, May 7, 2015 at 5:11 PM, Charles Z Henry czhenry@gmail.com wrote:
Hi Mario,
I know how to solve this problem, but I'm only part-way there. I have a similar problem, rejecting feedback when processing acoustic instruments. I'm sorry I didn't post what I've got sooner--it needs some more work. I can't really tell if I'll be finished with a decent patch soon.
I'm almost through with the derivation in high-level symbolic math, so you can follow along with the patch and work on it too.
See the latency tester first. It plays noise from to [dac~] and pics up the signal from [adc~]. If you want to get a really clean graph for measuring your system latency, do a loopback test with the cable running from output to input on your audio interface. It should run reasonably well over loudspeakers and will tell you the round trip latency. There are two sliders where you can adjust delay to the input (mic) signal or the output (speaker) signal.
Next, there's the spectrum analyzer patch. Open the spect-2048_test.pd patch, then right click on [spect-2048] and open it. This solves the spectrum analysis problem on block size 2048. The result is just written to some tables in [spect-2048], for you to visually see it.
When it's set to run on block size 16384, it needs to have about 90-100 ms of latency (don't even try opening it, without the audio buffer length turned up).
Once you've captured an impulse with the spectrum analyzer as it is now, you'd have to save the impulse to a file and then re-open with your runtime patch (and lower latency settings). The patch spect-16384_test.pd patch does a better job of calculating the impulse response than the 2048 block size patch.
I'd like to have reduced latency to run the spectrum analyzer patch with larger block sizes (for which I need a different [xcov~] that works well on overlapped block sizes). The syntax of [xcov~] is already so ugly and will get worse with overlapped blocks.
Also, the spectrum analyzer math needs a little work. More to come...
Chuck
On May 7, 2015 7:15 AM, "Mario Mey" mariomey@gmail.com wrote:
Yes, as I said in the mail, not only the the time of the delay is very very short that it would be very difficult for me to find "at hand", also, as you say, this time would be variable. My system don't work... I knew it.
I took a look to some pages (*), but I only understand... the title of the mails and the PDF filename. The texts are chinese for me. I didn't know that echo cancellation is, as Peter Parker said, "quite a substantial topic"... and it is out of my hands to try different EC systems. No time, no knowledge, no time... no time.
Also, there's http://grh.mur.at/software/adaptive.html... but I don't know how to test it.
I made a diagram of my system... if anyone knows which echo cancellation system would work for me... I would reallly appreciate it.
Thank you.
(*) http://comments.gmane.org/gmane.comp.multimedia.puredata.general/63658
https://ccrma.stanford.edu/~eberdahl/Projects/FeedbackCancellation/FeedbackC... Peter Parker sent me this link, but I have to pay to see it: http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=1146062.
El 04/05/15 a las 13:48, Spencer Russell escibió:
You're not likely to get very good results with just a single delay. The way these systems normally work (e.g. on phones and conferencing systems) is to continuously estimate the transfer function between the speaker and the microphone (impulse response in the time domain). Then you take the received signal, convolve it with the inverse of the impulse response, and mix it into the signal you receive from the microphone.
The diagram at the bottom of this page[1] is a pretty good one to understand what's going on.
-s
http://www.speex.org/docs/manual/speex-manual/node4.html
On Mon, May 4, 2015, at 10:18 AM, Mario Mey wrote:
Hi, there. 2 years ago, I created this thread on the forum: http://forum.pdpatchrepo.info/topic/7340/echo-supressor. After talking about dBs, I thought I acchived what I was looking for. But no. Nor close.
My last non-replied post on september 2014 says:
Bringing back to life an old thread... I tried to make this patch work LIVE... and it is not as thought.
To make this patch work (*1), I should find the perfect delay to suppress the audio. Using 44100, the delay time can vary in 0.0226ms. So... it's almost impossible to get that. I only acchieve a flanger, nothing else.
In the first Patch-Circle done in Argentina last Saturday, I talked with Pablo E. Riera about this and he told me about the adaptive external and the echo cancellation example... but there is no example like that (only intereference cancellation).
I will explain again (*2): the operator has a microphone (using one channel of input) and it sounds in a LCD (far away) (using one channel of output). Above the LCD I have a mic (second channel of input). This mic takes what the audience says and send to the operator headphones (second channel of output). The operator hears her own voice (with a delay). I want to supress this "return". It doesn't matter the latency, I just don't want the operator to hear her/him voice.
I would need something that learn from the first microphone and get rid of what the second mic takes.
I hope being clear in my explanation. Thanks in advace.
(*1): the patch is in the thread. It does something like this: it takes the microphone (channel one), invert the wave (by using [*~ -1]), it uses a delayline to synch (because of the latency) and mix with the output. The intention is, obviously, suppress the microphone after been listened by the second microphone.
[adc~] |
| [delwrite~ echo-supressor-A 1000] | [delwrite~ echo-supressor-B 1000][Hslider] # to synch. It depends on latency (it would have to be exact, here is the problem!) | [vd~ echo_supressor-A] | [*~ -1] | | [vd~ echo_supressor-B 1] | / [+~] | [dac~]
(*2): I just updated the explanation
How can I acchieve an echo suppresor? Is this possible? If I'm not clear in my explanation, please tell me. Thanks. _______________________________________________ Pd-list@lists.iem.at mailing list UNSUBSCRIBE and account-management -> http://lists.puredata.info/listinfo/pd-list
Pd-list@lists.iem.at mailing list UNSUBSCRIBE and account-management -> http://lists.puredata.info/listinfo/pd-list
Pd-list@lists.iem.at mailing list UNSUBSCRIBE and account-management -> http://lists.puredata.info/listinfo/pd-list
-
Charles Z Henry -
Mario Mey -
Peter P. -
Spencer Russell