Calculate MFCC features for the audio in a given file Args: audio_filename: file name of the audio Returns: feature_vect

Calculate spectrogram for the audio file Args: audio_filename: audio file name Returns: log spectrogram values

Traditional feedforward layer: multiply on weight matrix, add bias vector and apply activation function Args: x

Replace every "n" values by their average Args: arr: input array n: number of elements to average on Returns: resul

Extract all 5 prosodic features Args: audio_filename: file name for the audio to be used Returns: pros_feature: ene

Return result of a net after n layers or n-1 layer (if is_target is true) This function will be used for the layer-wise pretraining of the

Create a dataset Args: name: dataset: 'train' or 'test' or 'dev nodes: markers used in motion caption Returns:

Train or restore a neural network Args: train_data: training dataset normalized to the values [-1,1] dev_data: de

Read and preprocess the motion dataset Args: data_dir: a directory with the dataset Return: Y_train:

Reads a file from CMU MoCap dataset in BVH format Returns: sequence [sequence_length,frame_size] - local chanells transfor

Changing the shape of the dataset array to correspond to the frame dimentionality Args: dataset: an array of the dataset Return:

Save body keypoint positions into file Args: out_dir: output directory gesture_name: basename of the output file

Compute speed between adjacent frames Args: data: array containing joint positions of gesture dim: gesture

Construct a TensorFlow graph for the AutoEncoding network Args: input_seq_pl: tf placeholder for input data: size [batch_size,

Extract features from a given pair of audio and motion files Args: audio_filename: file name for an audio file (.wav) gest

Calculate numerical derivative (by FDM) of a 1d array Args: x: input space x f: Function of x Returns: der: numerica

Compute Hellinger distance between two histograms Args: hist1: first histogram hist2: second histogram of the

Reconstruction error. Square of the RMSE Args: output: tensor of net output target: tensor of net we are trying to reconstruct

Write computed histogram to CSV Args: lines: list of strings to be written out_dir: output directory

Write computed measure to CSV Args: lines: list of strings to be written out_dir: output directory me

Apply simple moving average filter to a gesture Args: pos_array: body keypoint positions to filter winlen: averagi

Read test sequence in c3d format and write it into the binart file Args: input_file_name: the name of the input file

Helper to create an initialized Variable with weight decay. Note that the Variable is initialized with a truncated normal distribution.

Add Gaussian noise to the data Args: x - input vector variance_multiplier - coe

Apply one euro filter to a gesture Original implementation can be downloaded from http://cristal.univ-lille.fr/~casiez/1euro/ Ar

Load bvh hierarchy nodes Args: filename: name of the hierarchy file Returns: nodes: bvh hierarchy nodes

Create hierarchy nodes: an array of markers used in the motion capture Args: hierarchy: bvh file read in a structure Returns:

Create test sequences Args: nodes: markers used in motion caption dataset: dataset name ('train', 'test' or 'dev') R

Compute Hellinger distance between two histograms Args: hist1: first histogram hist2: second histogram of the

Pretrain AutoEncoding neural network in a layer-wise way Args: nn: neural network to be trained trainers: opt

Load bvh motion frames Args: filename: bvh filename Returns: frames: list of bvh frames

Pad array of features in order to be able to take context at each time-frame We pad N_CONTEXT / 2 frames before and after the signal by the f

Predict human gesture based on the speech Args: model_name: name of the Keras model to be used input_file: file name of the au

Reads 3 datasets: "Train","Dev" and "Eval" from the CMU MoCap dataset in bvh format And write them in the binary format.

Read motion capture's body joint names from file Args: filename: file name to read Returns: joint_names: list o

Read motion capture's body joint names from file Args: filename: file name to read Returns: joint_names: list o

Read motion capture's body joint names from file Args: filename: file name to read Returns: joint_names: list o

Read motion capture's body joint names from file Args: filename: file name to read Returns: joint_names: list o

Read motion capture's body joint names from file Args: filename: file name to read Returns: joint_names: list o

Reads all 3 datasets from CMU MoCap dataset in C3D format Args: train_dir - address to the train, dev and eval datasets

Remove velocity values from raw prediction data Args: data: array containing both position and velocity values dim:

Remove velocity values from raw prediction data Args: data: array containing both position and velocity values dim:

Transform vectors of the human motion from the joint angles to the absolute positions Args: frames: human motion in the join angles s

Convert bvh frames to body keypoint positions Args: frames: bvh frames nodes: bvh hierarchy nodes Retur

Save body keypoint positions into file Args: out_dir: output directory gesture_name: basename of the output file

Train a neural network to take speech as input and produce gesture as an output Args: model_file: file to store the model Retur

Compute Average Position Error (APE) Args: original: array containing joint positions of original gesture predicted:

Compute Mean Absolute Error (MAE) Args: original: array containing joint positions of original gesture predicted: ar

Get AutoEncoder tf variable Returns the specified variable created by this object. Names are weights#, biases#, biases#_out, weights#

DataInfo initializer Args: data_sigma: variance in the dataset train_shape: dimensionality of the train dataset

DAE initializer Args: shape: list of ints specifying num input, hidden1 units,...hidden_n units,

Store a TensorFlow variable NOTE: Don't call this explicitly. It should be used only internally when setting up variables.

Annotate barplot with p-values. :param num1: number of left bar to put bracket over :param num2: number of right bar to put bracket over

Calculate F0 contour of a given speech file Args: audio_filename: address of a speech file Returns: F0 contour in a log scal

Compute acceleration between adjacent frames Args: data: array containing joint positions of gesture dim:

Compute acceleration between adjacent frames Args: data: array containing joint positions of gesture dim:

Compute jerk between adjacent frames Args: data: array containing joint positions of gesture dim: gesture

Compute velocity between adjacent frames Args: data: array containing joint positions of gesture dim: gest