Support for providing media stream audio track to speech recognition

Source/modules/speech/SpeechRecognitionController.h

Index: Source/modules/speech/SpeechRecognitionController.h
diff --git a/Source/modules/speech/SpeechRecognitionController.h b/Source/modules/speech/SpeechRecognitionController.h
index 843d065f8ae673cff9171ddf449978fc59238a50..a4cca6877bc339153496b8ad5c7d7400068a1262 100644
--- a/Source/modules/speech/SpeechRecognitionController.h
+++ b/Source/modules/speech/SpeechRecognitionController.h
@@ -27,6 +27,7 @@
 #define SpeechRecognitionController_h
 
 #include "core/page/Page.h"
+#include "modules/mediastream/MediaStreamTrack.h"
 #include "modules/speech/SpeechRecognitionClient.h"
 #include "wtf/PassOwnPtr.h"
 
@@ -37,9 +38,9 @@ class SpeechRecognitionController FINAL : public NoBaseWillBeGarbageCollectedFin
 public:
     virtual ~SpeechRecognitionController();
 
-    void start(SpeechRecognition* recognition, const SpeechGrammarList* grammars, const String& lang, bool continuous, bool interimResults, unsigned long maxAlternatives)
+    void start(SpeechRecognition* recognition, const SpeechGrammarList* grammars, const String& lang, bool continuous, bool interimResults, unsigned long maxAlternatives, MediaStreamTrack* audioTrack)

[Mike West, 2014/09/17 10:20:27]

Hrm. The various '::start()' methods are getting a bit unwieldy. Perhaps it
would make sense to create an initialization object (struct) that you could pass
in here, rather than relying on memory and the compiler's type checking to
ensure that everything is passed in and piped through in the right order?

[burnik, 2014/09/17 11:17:20]

Could you take a look at the previous patchset? I've added attach/detach which
at least partially avoids this.

Controller and Client just pass thru data. No real logic until reaching to the
public web interface.

On 2014/09/17 10:20:27, Mike West wrote:

[Mike West, 2014/09/17 11:30:46]

I like this approach more than the separate attach/detach steps. This seems like
the right direction to go, the argument list just seems to be getting pretty
long. I certainly won't insist on refactoring all the calls to ::start in this
CL, but I do think it's worth moving to a one-parameter model in the future to
keep things simple.
Yup. My claim is simply that bringing those parameters out into a separate
object a) makes it clear exactly what bits and pieces are required, and b) makes
it simpler to understand what needs to be passed around in order to kick off
recognition.

     {
-        m_client->start(recognition, grammars, lang, continuous, interimResults, maxAlternatives);
+        m_client->start(recognition, grammars, lang, continuous, interimResults, maxAlternatives, audioTrack);
     }
 
     void stop(SpeechRecognition* recognition) { m_client->stop(recognition); }