Project in several phases to ultimately built an app that translate from human to cats and cats to human using deep learning algorithm For all cat lovers by Lea B, Teo B and Herve B

web site link: here

Development wiki link: here

--> Current Status

Help me to collect cat sounds either by recording your cat and droping the file in the drop box below or capturing sounds from Youtube videos using online mp3fy here or any recording app on mobile (see how to name the file below)

• Phase 1: Collection of cat sound samples in a specific format with annotation of the cat behaviour following a prescribed method for each predesignated designed behavours (eg hungry, don't bother me, etc.)
• Phase 2: Data manipulation and preparation for deep learning
• Phase 3: Machine learning on initial sample (this could branch out depending on machine learning algorith used)
• Phase 4: App development to apply the learned model to real situation and use language recognition to speak cat language

Drop your sound files here according to instructions on link above:

This task is probably the most challenging task.

This requires either volunteers to record their own cats, which could be fun, though time consuming or screening online resources such as Youtube and then post processing the audio files. In any case wave files need to be collected.

Submission of cat sounds to: CatSounds Drop Box

Tools to collect cats sound : mp3fy here or any recording app on mobile

Naming of raw cat sounds files: Cat sounds should be named as follows: "SoundClass_Sex_Age_Name_FistNameOfWhoIsSubmitting_Country_Source"

where,

• SoundClass is one of the following:

• Sex is either M for male of F for female (if known)
• Age is age of cat (approximately if known, eg 3m or 2Y, etc.) (if known)
• Name is name of cat (if known)
• FirstNameOfWhoIsSubmitting is name of who is sending the file
• *Country is country where the cat is. (this will be used to find out if cats in different countries understand each others) (if known)
• Source should be Web or Recorded

Hear the cats and what they say at: Link here

After a bit of research, it is proposed to follow the work done by JH Lee Domestic Cat Sound Classification Using Transfer Learning. The following site shows examples of sound within each classis used as a guide for classification of site The research proposes to classify cats sounds in 10 classes as follows:

Data filtering [HOLD: not decided yet]

It is curretly proposed to use Librosa library in python for MFCCS feature extractions

It is currently proposed to use Supervised Training Convulational Neural Networks CNN with Tensorflow and Keras given its accuracy and portability. The model will be developed in Python hand the model will be saved for further used in other languages. The target accuracy is 70%.

Simple app will be developed using MIT App Inventor to record cat and translate to human and vice versa.