The average speech probability for the higher band is multiplied by the quotient of the process and analyze powers, to avoid thinking that suppressed echo is speech. In order to do this both magnitudes, alanyze and process, needed to be stored. This also was used to calculate different previous STSA estimates for analyze and process.
This CL was tested on two long team member recordings (bjornv and kwiberg) and the noisiest (5) recordings from the QA set.
BUG=webrtc:3763
R=andrew@webrtc.org, bjornv@webrtc.org
Review URL: https://webrtc-codereview.appspot.com/23799004
git-svn-id: http://webrtc.googlecode.com/svn/trunk@7437 4adac7df-926f-26a2-2b94-8c16560cd09d
The implementation of WEBRTC_SPL_RSHIFT_W16 is simply >>. This CL removes the macro usage in audio_processing and signal_processing.
Affected components:
* aecm
* agc
* nsx
Indirectly affecting (through signal_processing changes)
* codecs/cng
* codecs/isac/fix
* codecs/isac/main
BUG=3348,3353
TESTED=locally on Linux and trybots
R=kwiberg@webrtc.org
Review URL: https://webrtc-codereview.appspot.com/28699005
git-svn-id: http://webrtc.googlecode.com/svn/trunk@7432 4adac7df-926f-26a2-2b94-8c16560cd09d
> Remove the different block lengths in ns_core
>
> This CL has bit-exact output.
>
> What it does:
> * Remove the blockLen10Ms, as it is hardcoded to be equal to blockLen.
> * This makes outLen to be always zero, so it can be removed too.
> * It also avoids the need to have an outBuf, because it is not used, so it is also removed
> * Replaced blockLen10Ms by blockLen everywhere, since they were hardcoded to be equal.
> * We don't need to check if outLen is zero, because it always is, so it was removed.
> * Of course, the outBuf needs no initial set or copying around, because it is not used.
>
> BUG=webrtc:3811
> R=bjornv@webrtc.org, kwiberg@webrtc.org
>
> Review URL: https://webrtc-codereview.appspot.com/30539004TBR=aluebs@webrtc.org
Review URL: https://webrtc-codereview.appspot.com/26629004
git-svn-id: http://webrtc.googlecode.com/svn/trunk@7306 4adac7df-926f-26a2-2b94-8c16560cd09d
This CL has bit-exact output.
What it does:
* Remove the blockLen10Ms, as it is hardcoded to be equal to blockLen.
* This makes outLen to be always zero, so it can be removed too.
* It also avoids the need to have an outBuf, because it is not used, so it is also removed
* Replaced blockLen10Ms by blockLen everywhere, since they were hardcoded to be equal.
* We don't need to check if outLen is zero, because it always is, so it was removed.
* Of course, the outBuf needs no initial set or copying around, because it is not used.
BUG=webrtc:3811
R=bjornv@webrtc.org, kwiberg@webrtc.org
Review URL: https://webrtc-codereview.appspot.com/30539004
git-svn-id: http://webrtc.googlecode.com/svn/trunk@7297 4adac7df-926f-26a2-2b94-8c16560cd09d
In practice, we have been doing this since time immemorial, but have
relied on the user to do the downmixing (first voice engine then
Chromium). It's more logical for this burden to fall on AudioProcessing,
however, who can be expected to know that this is a reasonable approach
for AEC. Permitting two render channels results in running two AECs
serially.
Critically, in my recent change to have Chromium adopt the float
interface:
https://codereview.chromium.org/420603004
I removed the downmixing by Chromium, forgetting that we hadn't yet
enabled this feature in AudioProcessing. This corrects that oversight.
The change in paths hit by production users is very minor. As commented
it required adding downmixing to the int16_t path to satisfy
bit-exactness tests.
For reference, find the ApmTest.Process errors here:
https://paste.googleplex.com/6372007910309888
BUG=webrtc:3853
TESTED=listened to the files output from the Process test, and verified
that they sound as expected: higher echo while the AEC is adapting, but
afterwards very close.
R=aluebs@webrtc.org, bjornv@webrtc.org, kwiberg@webrtc.org
Review URL: https://webrtc-codereview.appspot.com/31459004
git-svn-id: http://webrtc.googlecode.com/svn/trunk@7292 4adac7df-926f-26a2-2b94-8c16560cd09d
This attenuates the noise pumping generated from the NS adapting to the AEC comfort noise.
When there is echo present the AEC suppresses it and adds comfort noise. This is underestimated on purpose to avoid adding more than the original background noise. The NS has to be called after the AEC, because every non-linear processing before it can ruin its performance. Therefore the noise estimation can adapt to this comfort noise, making it less aggressive and generating noise pumping.
By putting the noise estimation analysis stage from the NS before the AEC, this effect can be avoided. This has been tested manually on recordings where noise pumping was present: Two long recordings done in our team by bjornv and kwiberg plus the most noisy (5) recordings in the QA set.
On the other hand, one risk of doing this is to not adapt to the comfort noise and therefore suppress too much. As verified in the tested files, this is not a problem in practice.
BUG=webrtc:3763
R=andrew@webrtc.org, bjornv@webrtc.org
Review URL: https://webrtc-codereview.appspot.com/24679004
git-svn-id: http://webrtc.googlecode.com/svn/trunk@7289 4adac7df-926f-26a2-2b94-8c16560cd09d
webrtc did not build if AGC_DEBUG was turned on. This CL fixes that. Has no impact on performance since it is development/debug code.
* Name change to WEBRT_AGC_DEBUG_DUMP
* Added build flag agc_debug_dump to .gypi
* Added missing "%d" in printf at two places
* Some line length related style changes
Tested audioproc and modules_unittests with GYP_DEFINES=agc_debug_dump=1 webrtc/build/gyp_webrtc
BUG=N/A
TESTED=locally and trybots
R=aluebs@webrtc.org, kwiberg@webrtc.org
Review URL: https://webrtc-codereview.appspot.com/31429004
git-svn-id: http://webrtc.googlecode.com/svn/trunk@7271 4adac7df-926f-26a2-2b94-8c16560cd09d
The reason why ApmTest.Process breaks on Android is that two metrics over counts. I decided to add an offset and a different slack to the EXPECT_NEAR() calls that are affected. I think this is a reasonable approach since we have no more than two failing metrics. If any feature change that will make another metric fail, we should go back to the desk and find another way of solving this.
BUG=114
TESTED=locally on Nexus 7 and trybots
R=aluebs@webrtc.org, andrew@webrtc.org, kwiberg@webrtc.org
Review URL: https://webrtc-codereview.appspot.com/26509004
git-svn-id: http://webrtc.googlecode.com/svn/trunk@7268 4adac7df-926f-26a2-2b94-8c16560cd09d
Filled the empty analyze API, separating the noise estimation from the process API.
No formatting fixes or extra refactoring has been done, to make the review process easier.
This patch has been tested for bit-exactness over the whole QA set in every aggressiveness.
BUG=webrtc:3811
R=bjornv@webrtc.org
Review URL: https://webrtc-codereview.appspot.com/27549004
git-svn-id: http://webrtc.googlecode.com/svn/trunk@7243 4adac7df-926f-26a2-2b94-8c16560cd09d
These optimizations were originally committed in r6860, but reverted in r6861, since it broke a bitexactness test (ApmTest.Process) in modules_unittests. That test has now been updated in r7149, hence this CL now pass the test.
BUG=3767
TESTED=manually on linux and trybots
TBR=aluebs@webrtc.org
Review URL: https://webrtc-codereview.appspot.com/25539004
git-svn-id: http://webrtc.googlecode.com/svn/trunk@7189 4adac7df-926f-26a2-2b94-8c16560cd09d
When writing to wav files in the low level flag aec_debug_dump incorrect sample rates were used for recordings using rates from 32 kHz and above. This since internally inside the AEC we process the data using 16 kHz. Any upper band is processed and combined later on.
This CL adds the correct sample rate to the recording.
BUG=3359
TESTED=locally on 44.1 kHz recordings on Linux
R=aluebs@webrtc.org
Review URL: https://webrtc-codereview.appspot.com/23649004
git-svn-id: http://webrtc.googlecode.com/svn/trunk@7182 4adac7df-926f-26a2-2b94-8c16560cd09d
