Adds SpeechInputManagerImpl to handle requests from render views and return recognition events.

Adds SpeechInputManagerImpl to handle requests from render views and return recognition events.
This CL depends on http://codereview.chromium.org/3124009 going in first for the SpeechRecognizer class.
Also renamed SpeechInputManager::Listener to Delegate to be consistent with Chromium terminology.

TEST=manually, open a webpage containing an <input speech> tag and click on the speech button to test.
BUG=none

Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=56002

[bulach, 2010-08-11 10:55:12]

LGTM

looks fine, thanks! some nits and suggestions below:

http://codereview.chromium.org/3108007/diff/1/4
File chrome/browser/speech/speech_input_manager.cc (right):

http://codereview.chromium.org/3108007/diff/1/4#newcode16
chrome/browser/speech/speech_input_manager.cc:16:
SpeechInputManagerImpl(SpeechInputManagerDelegate* delegate);
explicit

http://codereview.chromium.org/3108007/diff/1/4#newcode50
chrome/browser/speech/speech_input_manager.cc:50:
recognizers_.begin()->second->StopRecording(true);
hmm, shouldn't this class outlive the recognizers, i.e., here it should check
that recognizers_.empty() ?

http://codereview.chromium.org/3108007/diff/1/4#newcode57
chrome/browser/speech/speech_input_manager.cc:57: return;
hmm, the DCHECK and the if are a bit contradictory.. I'd think that it's either
fine to start an existing render_view_id again (and then remove the check), or
it's not allowed, so remove the if?

http://codereview.chromium.org/3108007/diff/1/4#newcode70
chrome/browser/speech/speech_input_manager.cc:70: if
(recognizers_.find(render_view_id) != recognizers_.end()) {
as above..

http://codereview.chromium.org/3108007/diff/1/4#newcode76
chrome/browser/speech/speech_input_manager.cc:76:
DCHECK(recognizers_.find(render_view_id) != recognizers_.end());
as above

[Satish, 2010-08-11 11:27:30]

Thanks for the quick review. A couple of replies below.

http://codereview.chromium.org/3108007/diff/1/4
File chrome/browser/speech/speech_input_manager.cc (right):

http://codereview.chromium.org/3108007/diff/1/4#newcode50
chrome/browser/speech/speech_input_manager.cc:50:
recognizers_.begin()->second->StopRecording(true);
On 2010/08/11 10:55:13, bulach wrote:
> hmm, shouldn't this class outlive the recognizers, i.e., here it should check
> that recognizers_.empty() ?

Technically yes it should outlive. But in case there is a render view which
crashes or due to a bug doesn't call CancelRecognition before it closes, we'll
have some recognizers hanging in here. We need to call StopRecording explicitly
to stop the audio recorder and free up resources properly, hence I added this
here.

http://codereview.chromium.org/3108007/diff/1/4#newcode57
chrome/browser/speech/speech_input_manager.cc:57: return;
On 2010/08/11 10:55:13, bulach wrote:
> hmm, the DCHECK and the if are a bit contradictory.. I'd think that it's
either
> fine to start an existing render_view_id again (and then remove the check), or
> it's not allowed, so remove the if?

From what I know DCHECK only checks the condition in debug builds, and I felt it
was better to sanity check than let the code crash in a release build. I can
remove the check if you disagree, as there are pros and cons to both approaches.