Bug 1940906 - Add the unspokenPunctuation attribute to SpeechRecognition. r=media-playback-reviewers,webidl,alwu,smaug

Per spec, unspokenPunctuation defaults to false; the value is stored for
round-tripping but does not change recognition behaviour, since Gecko's
recognizer is LLM-based and only infers punctuation that was not spoken.

Spec: https://webaudio.github.io/web-speech-api/#dom-speechrecognition-unspokenpunctuation
WPT: covered by interfaces/speech-api.idl via idlharness.

Differential Revision: https://phabricator.services.mozilla.com/D311119

Author: padenot

dom/media/webspeech/recognition/SpeechRecognition.cpp

@@ -388,6 +388,14 @@ void SpeechRecognition::SetProcessLocally(bool aProcessLocally) {
   mProcessLocally = aProcessLocally;
 }
 
+bool SpeechRecognition::UnspokenPunctuation() const {
+  return mUnspokenPunctuation;
+}
+
+void SpeechRecognition::SetUnspokenPunctuation(bool aUnspokenPunctuation) {
+  mUnspokenPunctuation = aUnspokenPunctuation;
+}
+
 void SpeechRecognition::OnSetPhrases(SpeechRecognitionPhrase& aPhrase,
                                      uint32_t aIndex, ErrorResult& aRv) {
   // Note: The spec is unclear on whether dynamic updates during recognition

dom/media/webspeech/recognition/SpeechRecognition.h

@@ -63,6 +63,8 @@ class SpeechRecognitionInstallTransaction final {
   nsTArray<RefPtr<Promise>> mPromises;
 };
 
+// This implements the SpeechRecognition object in the content process, from the
+// Web Speech API: https://webaudio.github.io/web-speech-api/#speechrecognition
 class SpeechRecognition final : public DOMEventTargetHelper,
                                 public SupportsWeakPtr {
  public:
@@ -111,6 +113,9 @@ class SpeechRecognition final : public DOMEventTargetHelper,
   bool ProcessLocally() const;
   void SetProcessLocally(bool aProcessLocally);
 
+  bool UnspokenPunctuation() const;
+  void SetUnspokenPunctuation(bool aUnspokenPunctuation);
+
   // ObservableArray callbacks for phrases
   void OnSetPhrases(SpeechRecognitionPhrase& aPhrase, uint32_t aIndex,
                     ErrorResult& aRv);
@@ -265,6 +270,10 @@ class SpeechRecognition final : public DOMEventTargetHelper,
   bool mInterimResults;
   uint32_t mMaxAlternatives;
   bool mProcessLocally = false;
+  // Per spec, defaults to false. The value is stored for round-tripping but
+  // does not change recognition behaviour: Gecko's recognizer is LLM-based
+  // and only ever infers punctuation that was not spoken.
+  bool mUnspokenPunctuation = false;
   // The backend gets these at Start() time; spec is unclear on dynamic updates
   // Probably better as a SimpleMap or something so it's sparse
   // https://github.com/WebAudio/web-speech-api/issues/172

dom/webidl/SpeechRecognition.webidl

@@ -35,6 +35,7 @@ interface SpeechRecognition : EventTarget {
     [Throws]
     attribute boolean continuous;
     attribute boolean interimResults;
+    attribute boolean unspokenPunctuation;
     attribute unsigned long maxAlternatives;
 
     attribute boolean processLocally;

testing/web-platform/meta/speech-api/idlharness.https.window.js.ini

@@ -424,7 +424,8 @@
       if os == "android": FAIL
 
   [SpeechRecognition interface: attribute unspokenPunctuation]
-    expected: FAIL
-
+    expected:
+      if os == "android": FAIL
   [SpeechRecognition interface: new SpeechRecognition() must inherit property "unspokenPunctuation" with the proper type]
-    expected: FAIL
+    expected:
+      if os == "android": FAIL