These are mostly trivial changes and are separated out just to reduce the
diff on that change to the minimum possible.
Note explanatory comments on patch set 1.
BUG=none
TEST=none
Review URL: https://codereview.webrtc.org/1235643003
Cr-Commit-Position: refs/heads/master@{#9617}
Reduces musical noise with minimal impact on interferer suppression.
This also unifies the treatment of "mean bins". The "end" bin is now
inclusive in the mean range as with the "start" bin.
Corrects interpretation of quantile.
BUG=chromium:490477
R=aluebs@webrtc.org
Review URL: https://webrtc-codereview.appspot.com/50939004
Cr-Commit-Position: refs/heads/master@{#9317}
From 6kHz-6.5kHz to 3kHz-5kHz. Previous range had unreliable mask values, letting high frequencies from all directions through. The new range is wider and lower, which results in better estimates.
R=andrew@webrtc.org
Review URL: https://webrtc-codereview.appspot.com/47089004
Cr-Commit-Position: refs/heads/master@{#9213}
This adds functionality from audioproc_float. The geometry parsing code
is now shared from test_utils.h. I removed the "mic_spacing" flag from
audioproc_float because it's a redundancy that I suspect isn't very
useful.
Includes a cleanup of the audio_processing test utils. They're now
packaged in targets, with the protobuf-using ones split out to avoid
requiring users to depend on protobufs.
pcm_utils is no longer needed and removed.
The primary motivation for this CL is that AudioProcessing currently
doesn't support more than two channels and we'd like a way to pass
more channels to the beamformer.
R=aluebs@webrtc.org, mgraczyk@chromium.org
Review URL: https://webrtc-codereview.appspot.com/50899004
Cr-Commit-Position: refs/heads/master@{#9157}
* Don't use ConjugateDotProduct to calculate the norm.
* Only resize Matrix when needed.
This makes the Beamformer run in 93.6% the original time.
The error between the new and original output is really small and is caused by the new norm calculation.
R=andrew@webrtc.org
Review URL: https://webrtc-codereview.appspot.com/37339004
Cr-Commit-Position: refs/heads/master@{#8438}
git-svn-id: http://webrtc.googlecode.com/svn/trunk@8438 4adac7df-926f-26a2-2b94-8c16560cd09d
There are only 2 things that prevent the output to be bit-exact:
* The zero initialization of the postfilter_mask_ and high_pass_postfilter_mask_, which only afects the first blocks.
* The re-tuning of the target presence estimation, since only the bins between low_average_start_bin_ and high_average_end_bin_ are of interest.
This latter was not taken into account before.
R=andrew@webrtc.org
Review URL: https://webrtc-codereview.appspot.com/35139004
Cr-Commit-Position: refs/heads/master@{#8368}
git-svn-id: http://webrtc.googlecode.com/svn/trunk@8368 4adac7df-926f-26a2-2b94-8c16560cd09d
Now the ChannelBuffer has 2 separate arrays, one for the full-band data and one for the splitted one. The corresponding accessors are added to the ChannelBuffer.
This is done to avoid having to refresh the bands pointers in AudioBuffer. It will also allow us to have a general accessor like data()[band][channel][sample].
All the files using the ChannelBuffer needed to be re-factored.
Tested with modules_unittests, common_audio_unittests, audioproc, audioproc_f, voe_cmd_test.
R=andrew@webrtc.org, kwiberg@webrtc.org
Review URL: https://webrtc-codereview.appspot.com/36999004
Cr-Commit-Position: refs/heads/master@{#8318}
git-svn-id: http://webrtc.googlecode.com/svn/trunk@8318 4adac7df-926f-26a2-2b94-8c16560cd09d
Take the 50% quantile of the mask and compare it to certain threshold to determine if the desired signal is present. A hold is applied to avoid fast switching between states.
is_signal_present_ has been plotted and looks as expected. The AGC adaptation sounds promising, specially for the cases when the speaker fades in and out from the beam direction.
R=andrew@webrtc.org
Review URL: https://webrtc-codereview.appspot.com/28329005
git-svn-id: http://webrtc.googlecode.com/svn/trunk@8078 4adac7df-926f-26a2-2b94-8c16560cd09d
