Speech input: Do environment estimation and detect the no-speech case.

Give time for environment estimation and detect the no-speech case.
Environment estimation helps establish a baseline against which the endpointer
compares to find beginning and end of speech. We now do this estimation for a
few hundred milliseconds before showing the speech UI asking user to speak.
Using this updated endpointer we also detect if no speech was heard for many
seconds and if so inform the delegate as an error. In a future CL we'll allow
the user to easily retry recognition in these error cases.

BUG=53598
TEST=unit_tests --gtest_filter=SpeechRecognizerTests.*

Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=58930

[joth, 2010-09-09 10:51:37]

LGTM

couple small suggestions...

http://codereview.chromium.org/3341020/diff/2001/3003
File chrome/browser/speech/speech_input_manager.cc (right):

http://codereview.chromium.org/3341020/diff/2001/3003#newcode192
chrome/browser/speech/speech_input_manager.cc:192: const SpeechInputRequest&
request = requests_.find(caller_id)->second;
maybe easier as const SpeechInputRequest& request = requests_[caller_id];
also I'd be tempted to include the DCHECK(HasPendingRequest(caller_id)); here
(i.e. on line immediately proceeding this)

http://codereview.chromium.org/3341020/diff/2001/3006
File chrome/browser/speech/speech_recognizer_unittest.cc (right):

http://codereview.chromium.org/3341020/diff/2001/3006#newcode91
chrome/browser/speech/speech_recognizer_unittest.cc:91: int
audio_packet_length_bytes_;
would a vector<uint8> work better than these 2 members? (.size() will give you
the length)

http://codereview.chromium.org/3341020/diff/2001/3006#newcode209
chrome/browser/speech/speech_recognizer_unittest.cc:209: audio_packet_.get()[i]
= 0;
with vector, audio_packet_.resize(desired_length, 0) will do the job (you can
even drop the 0 as this is the default)

http://codereview.chromium.org/3341020/diff/2001/3006#newcode237
chrome/browser/speech/speech_recognizer_unittest.cc:237: audio_packet_.get()[i]
= 0;
ditto

http://codereview.chromium.org/3341020/diff/2001/3006#newcode243
chrome/browser/speech/speech_recognizer_unittest.cc:243: audio_packet_.get()[i]
= static_cast<uint8>(i);
would it be more representative to put a sine wave in here or something?
http://src.chromium.org/viewvc/chrome/trunk/src/media/audio/simple_sources.cc
has an example.

[Satish, 2010-09-09 11:30:19]

Addressed all comments with one reply below, will check in if you are ok with
it.

http://codereview.chromium.org/3341020/diff/2001/3006
File chrome/browser/speech/speech_recognizer_unittest.cc (right):

http://codereview.chromium.org/3341020/diff/2001/3006#newcode243
chrome/browser/speech/speech_recognizer_unittest.cc:243: audio_packet_.get()[i]
= static_cast<uint8>(i);
On 2010/09/09 10:51:37, joth wrote:
> would it be more representative to put a sine wave in here or something?
> http://src.chromium.org/viewvc/chrome/trunk/src/media/audio/simple_sources.cc
> has an example.

I thought of that earlier but a sine wave isn't exactly representing speech
either. So it would be swapping one artificial sequence for another, so I went
with this simpler for loop.

[joth, 2010-09-09 13:31:00]

LGTM.
could comment that it's approximating a 60Hz saw-tooth waveform, or just
leave as is..


On 9 September 2010 12:30, <satish@chromium.org> wrote:

> Addressed all comments with one reply below, will check in if you are ok
> with
> it.
>
>
>
> http://codereview.chromium.org/3341020/diff/2001/3006
> File chrome/browser/speech/speech_recognizer_unittest.cc (right):
>
> http://codereview.chromium.org/3341020/diff/2001/3006#newcode243
> chrome/browser/speech/speech_recognizer_unittest.cc:243:
> audio_packet_.get()[i] = static_cast<uint8>(i);
> On 2010/09/09 10:51:37, joth wrote:
>
>> would it be more representative to put a sine wave in here or
>>
> something?
>
>
> http://src.chromium.org/viewvc/chrome/trunk/src/media/audio/simple_sources.cc
>
>> has an example.
>>
>
> I thought of that earlier but a sine wave isn't exactly representing
> speech either. So it would be swapping one artificial sequence for
> another, so I went with this simpler for loop.
>
>
> http://codereview.chromium.org/3341020/show
>