diff options
Diffstat (limited to 'dom/media/webspeech/synth/nsSpeechTask.h')
-rw-r--r-- | dom/media/webspeech/synth/nsSpeechTask.h | 128 |
1 files changed, 128 insertions, 0 deletions
diff --git a/dom/media/webspeech/synth/nsSpeechTask.h b/dom/media/webspeech/synth/nsSpeechTask.h new file mode 100644 index 0000000000..fc121cf8f1 --- /dev/null +++ b/dom/media/webspeech/synth/nsSpeechTask.h @@ -0,0 +1,128 @@ +/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */ +/* vim:set ts=2 sw=2 sts=2 et cindent: */ +/* This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this + * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#ifndef mozilla_dom_nsSpeechTask_h +#define mozilla_dom_nsSpeechTask_h + +#include "SpeechSynthesisUtterance.h" +#include "AudioChannelAgent.h" +#include "nsISpeechService.h" +#include "nsWeakReference.h" + +namespace mozilla { + +class SharedBuffer; + +namespace dom { + +class SpeechSynthesisUtterance; +class SpeechSynthesis; + +class nsSpeechTask : public nsISpeechTask, + public nsIAudioChannelAgentCallback, + public nsSupportsWeakReference { + public: + NS_DECL_CYCLE_COLLECTING_ISUPPORTS + NS_DECL_CYCLE_COLLECTION_CLASS_AMBIGUOUS(nsSpeechTask, nsISpeechTask) + + NS_DECL_NSISPEECHTASK + NS_DECL_NSIAUDIOCHANNELAGENTCALLBACK + + explicit nsSpeechTask(SpeechSynthesisUtterance* aUtterance, + bool aShouldResistFingerprinting); + nsSpeechTask(float aVolume, const nsAString& aText, + bool aShouldResistFingerprinting); + + virtual void Pause(); + + virtual void Resume(); + + virtual void Cancel(); + + virtual void ForceEnd(); + + void SetSpeechSynthesis(SpeechSynthesis* aSpeechSynthesis); + + void Init(); + + void SetChosenVoiceURI(const nsAString& aUri); + + virtual void SetAudioOutputVolume(float aVolume); + + void ForceError(float aElapsedTime, uint32_t aCharIndex); + + bool IsPreCanceled() { return mPreCanceled; }; + + bool IsPrePaused() { return mPrePaused; } + + bool ShouldResistFingerprinting() { return mShouldResistFingerprinting; } + + enum { STATE_PENDING, STATE_SPEAKING, STATE_ENDED }; + + uint32_t GetState() const { return mState; } + + bool IsSpeaking() const { return mState == STATE_SPEAKING; } + + bool IsPending() const { return mState == STATE_PENDING; } + + protected: + virtual ~nsSpeechTask(); + + nsresult DispatchStartImpl(); + + virtual nsresult DispatchStartImpl(const nsAString& aUri); + + virtual nsresult DispatchEndImpl(float aElapsedTime, uint32_t aCharIndex); + + virtual nsresult DispatchPauseImpl(float aElapsedTime, uint32_t aCharIndex); + + virtual nsresult DispatchResumeImpl(float aElapsedTime, uint32_t aCharIndex); + + virtual nsresult DispatchErrorImpl(float aElapsedTime, uint32_t aCharIndex); + + virtual nsresult DispatchBoundaryImpl(const nsAString& aName, + float aElapsedTime, uint32_t aCharIndex, + uint32_t aCharLength, uint8_t argc); + + virtual nsresult DispatchMarkImpl(const nsAString& aName, float aElapsedTime, + uint32_t aCharIndex); + + RefPtr<SpeechSynthesisUtterance> mUtterance; + + float mVolume; + + nsString mText; + + bool mInited; + + bool mPrePaused; + + bool mPreCanceled; + + private: + void End(); + + void CreateAudioChannelAgent(); + + void DestroyAudioChannelAgent(); + + nsCOMPtr<nsISpeechTaskCallback> mCallback; + + RefPtr<mozilla::dom::AudioChannelAgent> mAudioChannelAgent; + + RefPtr<SpeechSynthesis> mSpeechSynthesis; + + nsString mChosenVoiceURI; + + bool mShouldResistFingerprinting; + + uint32_t mState; +}; + +} // namespace dom +} // namespace mozilla + +#endif |