path: root/chrome/browser/speech/speech_recognition_bubble.cc

// Copyright (c) 2012 The Chromium Authors. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.

#include "chrome/browser/speech/speech_recognition_bubble.h"

#include "base/bind.h"
#include "base/lazy_instance.h"
#include "base/message_loop/message_loop.h"
#include "content/public/browser/web_contents.h"
#include "content/public/browser/web_contents_view.h"
#include "grit/generated_resources.h"
#include "grit/theme_resources.h"
#include "ui/base/resource/resource_bundle.h"
#include "ui/gfx/canvas.h"
#include "ui/gfx/display.h"
#include "ui/gfx/image/image_skia_operations.h"
#include "ui/gfx/rect.h"
#include "ui/gfx/screen.h"

using content::WebContents;

namespace {

const color_utils::HSL kGrayscaleShift = { -1, 0, 0.6 };
const int kWarmingUpAnimationStartMs = 500;
const int kWarmingUpAnimationStepMs = 100;
const int kRecognizingAnimationStepMs = 100;

// A lazily initialized singleton to hold all the image used by the speech
// recognition bubbles and safely destroy them on exit.
class SpeechRecognitionBubbleImages {
 public:
  const std::vector<gfx::ImageSkia>& spinner() const { return spinner_; }
  const std::vector<gfx::ImageSkia>& warm_up() const { return warm_up_; }
  gfx::ImageSkia* mic_full() const { return mic_full_; }
  gfx::ImageSkia* mic_empty() const { return mic_empty_; }
  gfx::ImageSkia* mic_noise() const { return mic_noise_; }
  gfx::ImageSkia* mic_mask() const { return mic_mask_; }

 private:
  // Private constructor to enforce singleton.
  friend struct base::DefaultLazyInstanceTraits<SpeechRecognitionBubbleImages>;
  SpeechRecognitionBubbleImages();

  std::vector<gfx::ImageSkia> spinner_;  // Frames for the progress spinner.
  std::vector<gfx::ImageSkia> warm_up_;  // Frames for the warm up animation.

  // These images are owned by ResourceBundle and need not be destroyed.
  gfx::ImageSkia* mic_full_;  // Mic image with full volume.
  gfx::ImageSkia* mic_noise_;  // Mic image with full noise volume.
  gfx::ImageSkia* mic_empty_;  // Mic image with zero volume.
  gfx::ImageSkia* mic_mask_;  // Gradient mask used by the volume indicator.
};

SpeechRecognitionBubbleImages::SpeechRecognitionBubbleImages() {
  mic_empty_ = ui::ResourceBundle::GetSharedInstance().GetImageSkiaNamed(
      IDR_SPEECH_INPUT_MIC_EMPTY);
  mic_noise_ = ui::ResourceBundle::GetSharedInstance().GetImageSkiaNamed(
      IDR_SPEECH_INPUT_MIC_NOISE);
  mic_full_ = ui::ResourceBundle::GetSharedInstance().GetImageSkiaNamed(
      IDR_SPEECH_INPUT_MIC_FULL);
  mic_mask_ = ui::ResourceBundle::GetSharedInstance().GetImageSkiaNamed(
      IDR_SPEECH_INPUT_MIC_MASK);

  // The sprite image consists of all the animation frames put together in one
  // horizontal/wide image. Each animation frame is square in shape within the
  // sprite.
  const gfx::ImageSkia* spinner_image = ui::ResourceBundle::GetSharedInstance().
      GetImageSkiaNamed(IDR_SPEECH_INPUT_SPINNER);
  int frame_size = spinner_image->height();

  // When recording starts up, it may take a short while (few ms or even a
  // couple of seconds) before the audio device starts really capturing data.
  // This is more apparent on first use. To cover such cases we show a warming
  // up state in the bubble starting with a blank spinner image. If audio data
  // starts coming in within a couple hundred ms, we switch to the recording
  // UI and if it takes longer, we show the real warm up animation frames.
  // This reduces visual jank for the most part.
  SkBitmap empty_spinner;
  empty_spinner.setConfig(SkBitmap::kARGB_8888_Config, frame_size, frame_size);
  empty_spinner.allocPixels();
  empty_spinner.eraseRGB(255, 255, 255);
  // |empty_spinner| has solid color. Pixel doubling a solid color is ok.
  warm_up_.push_back(gfx::ImageSkia::CreateFrom1xBitmap(empty_spinner));

  for (gfx::Rect src_rect(frame_size, frame_size);
       src_rect.x() < spinner_image->width();
       src_rect.Offset(frame_size, 0)) {
    gfx::ImageSkia frame = gfx::ImageSkiaOperations::ExtractSubset(
        *spinner_image, src_rect);

    // The image created by ExtractSubset just points to the same pixels as
    // the original and adjusts rowBytes accordingly. However that doesn't
    // render properly and gets vertically squished in Linux due to a bug in
    // Skia. Until that gets fixed we work around by taking a real copy of it
    // below as the copied image has the correct rowBytes and renders fine.
    frame.EnsureRepsForSupportedScales();
    std::vector<gfx::ImageSkiaRep> image_reps = frame.image_reps();
    gfx::ImageSkia frame_copy;
    for (size_t i = 0; i < image_reps.size(); ++i) {
      const SkBitmap& copy_src = image_reps[i].sk_bitmap();
      SkBitmap copy_dst;
      copy_src.copyTo(&copy_dst, SkBitmap::kARGB_8888_Config);
      frame_copy.AddRepresentation(gfx::ImageSkiaRep(
          copy_dst, image_reps[i].scale()));
    }
    spinner_.push_back(frame_copy);

    // The warm up spinner animation is a gray scale version of the real one.
    warm_up_.push_back(gfx::ImageSkiaOperations::CreateHSLShiftedImage(
        frame_copy, kGrayscaleShift));
  }
}

base::LazyInstance<SpeechRecognitionBubbleImages> g_images =
    LAZY_INSTANCE_INITIALIZER;

}  // namespace

SpeechRecognitionBubble::FactoryMethod SpeechRecognitionBubble::factory_ = NULL;
const int SpeechRecognitionBubble::kBubbleTargetOffsetX = 10;

SpeechRecognitionBubble* SpeechRecognitionBubble::Create(
    WebContents* web_contents, Delegate* delegate,
    const gfx::Rect& element_rect) {
  if (factory_)
    return (*factory_)(web_contents, delegate, element_rect);

  // Has the tab already closed before bubble create request was processed?
  if (!web_contents)
    return NULL;

  return CreateNativeBubble(web_contents, delegate, element_rect);
}

SpeechRecognitionBubbleBase::SpeechRecognitionBubbleBase(
    WebContents* web_contents)
    : weak_factory_(this),
      animation_step_(0),
      display_mode_(DISPLAY_MODE_RECORDING),
      web_contents_(web_contents),
      scale_(1.0f) {
  gfx::NativeView view =
      web_contents_ ? web_contents_->GetView()->GetNativeView() : NULL;
  gfx::Screen* screen = gfx::Screen::GetScreenFor(view);
  gfx::Display display = screen->GetDisplayNearestWindow(view);
  scale_ = display.device_scale_factor();

  const gfx::ImageSkiaRep& rep =
      g_images.Get().mic_empty()->GetRepresentation(scale_);
  mic_image_.reset(new SkBitmap());
  mic_image_->setConfig(SkBitmap::kARGB_8888_Config,
      rep.pixel_width(), rep.pixel_height());
  mic_image_->allocPixels();

  buffer_image_.reset(new SkBitmap());
  buffer_image_->setConfig(SkBitmap::kARGB_8888_Config,
      rep.pixel_width(), rep.pixel_height());
  buffer_image_->allocPixels();
}

SpeechRecognitionBubbleBase::~SpeechRecognitionBubbleBase() {
  // This destructor is added to make sure members such as the scoped_ptr
  // get destroyed here and the derived classes don't have to care about such
  // member variables which they don't use.
}

void SpeechRecognitionBubbleBase::SetWarmUpMode() {
  weak_factory_.InvalidateWeakPtrs();
  display_mode_ = DISPLAY_MODE_WARM_UP;
  animation_step_ = 0;
  DoWarmingUpAnimationStep();
  UpdateLayout();
}

void SpeechRecognitionBubbleBase::DoWarmingUpAnimationStep() {
  SetImage(g_images.Get().warm_up()[animation_step_]);
  base::MessageLoop::current()->PostDelayedTask(
      FROM_HERE,
      base::Bind(&SpeechRecognitionBubbleBase::DoWarmingUpAnimationStep,
          weak_factory_.GetWeakPtr()),
      base::TimeDelta::FromMilliseconds(
          animation_step_ == 0 ? kWarmingUpAnimationStartMs
                               : kWarmingUpAnimationStepMs));
  if (++animation_step_ >= static_cast<int>(g_images.Get().warm_up().size()))
    animation_step_ = 1;  // Frame 0 is skipped during the animation.
}

void SpeechRecognitionBubbleBase::SetRecordingMode() {
  weak_factory_.InvalidateWeakPtrs();
  display_mode_ = DISPLAY_MODE_RECORDING;
  SetInputVolume(0, 0);
  UpdateLayout();
}

void SpeechRecognitionBubbleBase::SetRecognizingMode() {
  display_mode_ = DISPLAY_MODE_RECOGNIZING;
  animation_step_ = 0;
  DoRecognizingAnimationStep();
  UpdateLayout();
}

void SpeechRecognitionBubbleBase::DoRecognizingAnimationStep() {
  SetImage(g_images.Get().spinner()[animation_step_]);
  if (++animation_step_ >= static_cast<int>(g_images.Get().spinner().size()))
    animation_step_ = 0;
  base::MessageLoop::current()->PostDelayedTask(
      FROM_HERE,
      base::Bind(&SpeechRecognitionBubbleBase::DoRecognizingAnimationStep,
          weak_factory_.GetWeakPtr()),
      base::TimeDelta::FromMilliseconds(kRecognizingAnimationStepMs));
}

void SpeechRecognitionBubbleBase::SetMessage(const base::string16& text) {
  weak_factory_.InvalidateWeakPtrs();
  message_text_ = text;
  display_mode_ = DISPLAY_MODE_MESSAGE;
  UpdateLayout();
}

void SpeechRecognitionBubbleBase::DrawVolumeOverlay(SkCanvas* canvas,
                                                    const gfx::ImageSkia& image,
                                                    float volume) {
  buffer_image_->eraseARGB(0, 0, 0, 0);

  int width = mic_image_->width();
  int height = mic_image_->height();
  SkCanvas buffer_canvas(*buffer_image_);

  buffer_canvas.save();
  const int kVolumeSteps = 12;
  SkScalar clip_right =
      (((1.0f - volume) * (width * (kVolumeSteps + 1))) - width) / kVolumeSteps;
  buffer_canvas.clipRect(SkRect::MakeLTRB(0, 0,
      SkIntToScalar(width) - clip_right, SkIntToScalar(height)));
  buffer_canvas.drawBitmap(image.GetRepresentation(scale_).sk_bitmap(), 0, 0);
  buffer_canvas.restore();
  SkPaint multiply_paint;
  multiply_paint.setXfermodeMode(SkXfermode::kModulate_Mode);
  buffer_canvas.drawBitmap(
      g_images.Get().mic_mask()->GetRepresentation(scale_).sk_bitmap(),
      -clip_right, 0, &multiply_paint);

  canvas->drawBitmap(*buffer_image_.get(), 0, 0);
}

void SpeechRecognitionBubbleBase::SetInputVolume(float volume,
                                                 float noise_volume) {
  mic_image_->eraseARGB(0, 0, 0, 0);
  SkCanvas canvas(*mic_image_);

  // Draw the empty volume image first and the current volume image on top,
  // and then the noise volume image on top of both.
  canvas.drawBitmap(
      g_images.Get().mic_empty()->GetRepresentation(scale_).sk_bitmap(),
      0, 0);
  DrawVolumeOverlay(&canvas, *g_images.Get().mic_full(), volume);
  DrawVolumeOverlay(&canvas, *g_images.Get().mic_noise(), noise_volume);

  gfx::ImageSkia image(gfx::ImageSkiaRep(*mic_image_.get(), scale_));
  SetImage(image);
}

WebContents* SpeechRecognitionBubbleBase::GetWebContents() {
  return web_contents_;
}

void SpeechRecognitionBubbleBase::SetImage(const gfx::ImageSkia& image) {
  icon_image_ = image;
  UpdateImage();
}

gfx::ImageSkia SpeechRecognitionBubbleBase::icon_image() {
  return icon_image_;
}