Only update Intelligibility Enhancer gains every 10 chunks

Only update Intelligibility Enhancer gains every 10 chunks

This reduces its complexity by a factor of 2.7x total.
The mean error introduced by this is in the 6 different noise scenarios and 6 different speech signals tested is below -52dB.

Committed: https://crrev.com/e8f8f6037ca2b7f9c2f48cb65bb9e8b245acbddb
Cr-Commit-Position: refs/heads/master@{#13072}

[aluebs-webrtc, 2016-06-04 00:04:57]

aluebs@webrtc.org changed reviewers:
+ henrik.lundin@webrtc.org, peah@webrtc.org, turaj@webrtc.org

[turaj, 2016-06-04 00:48:04]

lgtm

[aluebs-webrtc, 2016-06-06 17:57:02]

Description was changed from

==========
Only update Intelligibility Enhancer gains every 10 chunks

This reduces its complexity by a factor of x2.8 total.
The error introduced by this is below -52dB.
==========

to

==========
Only update Intelligibility Enhancer gains every 10 chunks

This reduces its complexity by a factor of 2.7x total.
The error introduced by this is below -52dB.
==========

[peah-webrtc, 2016-06-06 21:47:04]

On 2016/06/06 17:57:02, aluebs-webrtc wrote:
> Description was changed from
> 
> ==========
> Only update Intelligibility Enhancer gains every 10 chunks
> 
> This reduces its complexity by a factor of x2.8 total.
> The error introduced by this is below -52dB.
> ==========
> 
> to
> 
> ==========
> Only update Intelligibility Enhancer gains every 10 chunks
> 
> This reduces its complexity by a factor of 2.7x total.
> The error introduced by this is below -52dB.
> ==========

lgtm, with the nit below.
I'm fine with the periodic updating, which for sure makes sense, but
I cannot see how you can ensure that the error introduced by this is below -52
dB. It should be fully scenario dependent, right? E.g., if there is a
significant background noise change between the 10th chunks, that will not be
visible in the IE output until 5 frames later, and the error could then be
higher until it has been recomputed at the 10 th frame.

Since you state it in the CL, please explain what you mean by that the error is
below -52 dB.

[aluebs-webrtc, 2016-06-06 21:55:29]

Description was changed from

==========
Only update Intelligibility Enhancer gains every 10 chunks

This reduces its complexity by a factor of 2.7x total.
The error introduced by this is below -52dB.
==========

to

==========
Only update Intelligibility Enhancer gains every 10 chunks

This reduces its complexity by a factor of 2.7x total.
The mean error introduced by this is in the 6 different noise scenarios and 6
different speech signals tested is below -52dB.
==========

[aluebs-webrtc, 2016-06-06 22:00:24]

On 2016/06/06 21:47:04, peah-webrtc wrote:
> On 2016/06/06 17:57:02, aluebs-webrtc wrote:
> > Description was changed from
> > 
> > ==========
> > Only update Intelligibility Enhancer gains every 10 chunks
> > 
> > This reduces its complexity by a factor of x2.8 total.
> > The error introduced by this is below -52dB.
> > ==========
> > 
> > to
> > 
> > ==========
> > Only update Intelligibility Enhancer gains every 10 chunks
> > 
> > This reduces its complexity by a factor of 2.7x total.
> > The error introduced by this is below -52dB.
> > ==========
> 
> lgtm, with the nit below.
> I'm fine with the periodic updating, which for sure makes sense, but
> I cannot see how you can ensure that the error introduced by this is below -52
> dB. It should be fully scenario dependent, right? E.g., if there is a
> significant background noise change between the 10th chunks, that will not be
> visible in the IE output until 5 frames later, and the error could then be
> higher until it has been recomputed at the 10 th frame.
> 
> Since you state it in the CL, please explain what you mean by that the error
is
> below -52 dB.

I updated the description to be clearer of how I estimated it. I didn't mean it
as "ensuring" that the error will be below -52dB, but more as an estimate of the
effect on some test scenarios.
Also, because of how the background noise is estimated using minimum statistics
and later smoothed in the power estimator, there can't be significant background
noise changes from one chunk to the other.

[aluebs-webrtc, 2016-06-07 22:48:49]

hlundin, do you have any additional comments or should I go ahead and land this?

[hlundin-webrtc, 2016-06-08 10:28:12]

lgtm

Sorry for the delay; holiday and offsite.

[aluebs-webrtc, 2016-06-08 15:17:32]

The CQ bit was checked by aluebs@webrtc.org

[commit-bot: I haz the power, 2016-06-08 15:17:35]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/2035213002/1

[commit-bot: I haz the power, 2016-06-08 16:53:22]

Description was changed from

==========
Only update Intelligibility Enhancer gains every 10 chunks

This reduces its complexity by a factor of 2.7x total.
The mean error introduced by this is in the 6 different noise scenarios and 6
different speech signals tested is below -52dB.
==========

to

==========
Only update Intelligibility Enhancer gains every 10 chunks

This reduces its complexity by a factor of 2.7x total.
The mean error introduced by this is in the 6 different noise scenarios and 6
different speech signals tested is below -52dB.
==========

[commit-bot: I haz the power, 2016-06-08 16:53:23]

Committed patchset #1 (id:1)

[commit-bot: I haz the power, 2016-06-08 16:53:30]

Description was changed from

==========
Only update Intelligibility Enhancer gains every 10 chunks

This reduces its complexity by a factor of 2.7x total.
The mean error introduced by this is in the 6 different noise scenarios and 6
different speech signals tested is below -52dB.
==========

to

==========
Only update Intelligibility Enhancer gains every 10 chunks

This reduces its complexity by a factor of 2.7x total.
The mean error introduced by this is in the 6 different noise scenarios and 6
different speech signals tested is below -52dB.

Committed: https://crrev.com/e8f8f6037ca2b7f9c2f48cb65bb9e8b245acbddb
Cr-Commit-Position: refs/heads/master@{#13072}
==========

[commit-bot: I haz the power, 2016-06-08 16:53:31]

Patchset 1 (id:??) landed as
https://crrev.com/e8f8f6037ca2b7f9c2f48cb65bb9e8b245acbddb
Cr-Commit-Position: refs/heads/master@{#13072}