-# Config file for FacialLandmarksForCubism
+# All coordinates used here are the ones used by the xdo library.
+# For a sense of how these work, run "xdotool getmouselocation".
-# The path of this config file should be passed to the constructor
-# of the FacialLandmarkDetector.
+# Milliseconds to sleep for between each update
+sleep_ms 5
-# Comments are lines that start with a '#' and are ignored by the parser.
-# Note that a line will be considered as a comment ONLY IF the '#' is the
-# very first character of the line, i.e. without any preceeding whitespace.
-
-
-## Section 1: dlib face detection and webcam parameters
-
-# Path to the dlib shape predictor trained dataset
-predictorPath ./shape_predictor_68_face_landmarks.dat
-
-# Value passed to the cv::VideoCapture() ctor
-cvVideoCaptureId 0
-
-# Number of milliseconds to wait after processing each video frame
-# This value controls the frame rate, but the actual frame period
-# is longer due to the time required to process each frame
-cvWaitKeyMs 5
-
-# If 1, show the webcam captured video on-screen; if 0, don't show
-showWebcamVideo 1
-
-# If 1, draw the detected facial landmarks on-screen; if 0, don't draw
-renderLandmarksOnVideo 1
-
-# If 1, laterally invert the image (create a mirror image); if 0, don't invert
-lateralInversion 1
-
-
-## Section 2: Cubism params calculation control
-#
-# These values control how the facial landmarks are translated into
-# parameters that control the Cubism model, and will vary from person
-# to person. The following values seem to work OK for my face, but
-# your milage may vary.
-
-# Section 2.0: Live2D automatic functionality
-# Set 1 to enable, 0 to disable.
-# If these are set, the automatic functionality in Live2D will be enabled.
-# Note: If you set auto blink, eye control will be disabled.
-autoBlink 0
-autoBreath 0
+# Automatic functionality in Live2D
+autoBlink 1
+autoBreath 1
randomMotion 0
-
-# Section 2.1: Face Y direction angle (head pointing up/down)
-# The Y angle is calculated mainly based on the angle formed
-# by the corners and the tip of the nose (hereafter referred
-# to as the "nose angle").
-
-# This applies an offset (in degrees).
-# If you have a webcam at the top of your monitor, then it is likely
-# that when you look at the centre of your monitor, the captured image
-# will have you looking downwards. This offset shifts the angle upwards,
-# so that the resulting avatar will still be looking straight ahead.
-faceYAngleCorrection 10
-
-# This is the baseline value for the nose angle (in radians) when looking
-# straight ahead...
-faceYAngleZeroValue 1.8
-
-# ... and this is when you are looking up...
-faceYAngleUpThreshold 1.3
-
-# ... and when looking down.
-faceYAngleDownThreshold 2.3
-
-# This is an additional multiplication factor applied per degree of rotation
-# in the X direction (left/right) - since the nose angle reduces when
-# turning your head left/right.
-faceYAngleXRotCorrection 0.15
-
-# This is the multiplication factor to reduce by when smiling or laughing -
-# the nose angle increases in such cases.
-faceYAngleSmileCorrection 0.075
-
-
-# Section 2.2: Eye control
-# This is mainly calculated based on the eye aspect ratio (eye height
-# divided by eye width). Note that currently an average of the values
-# of both eyes is applied - mainly due to two reasons: (1) the dlib
-# dataset I'm using fails to detect winks for me, and (2) if this is
-# not done, I frequently get asynchronous blinks which just looks ugly.
-
-# Maximum eye aspect ratio when the eye is closed
-eyeClosedThreshold 0.2
-
-# Minimum eye aspect ratio when the eye is open
-eyeOpenThreshold 0.25
-
-# Max eye aspect ratio to switch to a closed "smiley eye"
-eyeSmileEyeOpenThreshold 0.6
-
-# Min "mouth form" value to switch to a closed "smiley eye"
-# "Mouth form" is 1 when fully smiling / laughing, and 0 when normal
-eyeSmileMouthFormThreshold 0.75
-
-# Min "mouth open" value to switch to a closed "smiley eye"
-# "Mouth open" is 1 when fully open, and 0 when closed
-eyeSmileMouthOpenThreshold 0.5
-
-
-# Section 2.3: Mouth control
-# Two parameters are passed to Cubism to control the mouth:
-# - mouth form: Controls smiles / laughs
-# - mouth openness: How widely open the mouth is
-# Mouth form is calculated by the ratio between the mouth width
-# and the eye separation (distance between the two eyes).
-# Mouth openness is calculated by the ratio between the lip separation
-# (distance between upper and lower lips) and the mouth width.
-
-# Max mouth-width-to-eye-separation ratio to have a normal resting mouth
-mouthNormalThreshold 0.75
-
-# Min mouth-width-to-eye-separation ratio to have a fully smiling
-# or laughing mouth
-mouthSmileThreshold 1.0
-
-# Max lip-separation-to-mouth-width ratio to have a closed mouth
-mouthClosedThreshold 0.1
-
-# Min lip-separation-to-mouth-width ratio to have a fully opened mouth
-mouthOpenThreshold 0.4
-
-# Additional multiplication factor applied to the mouth openness parameter
-# when the mouth is fully smiling / laughing, since doing so increases
-# the mouth width
-mouthOpenLaughCorrection 0.2
-
-
-## Section 3: Filtering parameters
-# The facial landmark coordinates can be quite noisy, so I've applied
-# a simple moving average filter to reduce noise. More taps would mean
-# more samples to average over, hence smoother movements with less noise,
-# but it will also cause more lag between your movement and the movement
-# of the avatar, and quick movements (e.g. blinks) may be completely missed.
-
-faceXAngleNumTaps 11
-faceYAngleNumTaps 11
-faceZAngleNumTaps 11
-mouthFormNumTaps 3
-mouthOpenNumTaps 3
-leftEyeOpenNumTaps 3
-rightEyeOpenNumTaps 3
-
+useLipSync 1
+
+# Lip sync configurations
+
+# Gain to apply to audio volume.
+# (Linear gain multiplying audio RMS value)
+lipSyncGain 10
+
+# Cut-off volume for lip syncing.
+# If the volume is below this value, it will be forced to zero.
+# This can be useful if you don't have a good quality microphone
+# or are in a slightly noisy environment.
+# Note: this cut-off is applied *after* applying the gain.
+lipSyncCutOff 0.15
+
+# Audio buffer size. This is the window size over which we calculate
+# the "real-time" volume. A higher value will give a smoother
+# response, but will mean a higher latency.
+# The sampling rate is set to 44100 Hz so 4096 samples should
+# still have a low latency.
+audioBufSize 4096
+
+# Mouth form
+# Set this to 1 for a fully smiling mouth, 0 for a normal mouth,
+# and -1 for a frowning mouth, or anything in between.
+mouthForm 0
+
+# Screen number. Currently tracking is supported only in one screen.
+# If you have multiple screens, select the ID of the one you want to track.
+screen 0
+
+# The "middle" position, i.e. the coordinates of the cursor where
+# the Live2D model will be looking straight ahead.
+# For a 1920x1080 screen, {1600, 870} will be somewhere near the
+# bottom right corner.
+middle_x 1600
+middle_y 870
+
+# The bounding box. These are the limits of the coordinates where the
+# Live2D model will be looking 30 degrees to each side.
+top 0
+bottom 1079
+left 0
+right 1919
projects / mouse-tracker-for-cubism.git / blobdiff