Pet Sound Recognition System

What it does

A pet health identification system designed for pet owners who have pets at home but are unable to perceive the health status of their pets in time due to work and study.

Your inspiration

The topic "PetSound" was inspired by many cat and dog language translators on the market, and both members of the group were very fond of small animals.

How it works

After hearing the pet's meow, the large model classifier mobilizes the database to view the MFCC voiceprint features that match the meow, and then judge whether it is healthy or not.

Design process

The overall logic is very simple, and the division of labor between data collection and large model training is carried out at the same time. The data collection is manually identified by crawlers (using cat and dog speakers, of course, there are also inaccurate and unprofessional problems) and sorted, processed and visualized after data collection, and we chose the simpler MFCC voiceprint feature extraction. Cursor's AI was used to build the prototype for large model training, and Random Forest was selected as the support after horizontal comparison. Of course, this is our final version choice. At the beginning, the recognition accuracy of the version was only 33%, so we increased the accuracy to 81% by continuously expanding the database, changing the classifier model, and using MFCC voiceprint features and several core parameters to analyze the difference change, and added a variety of visualization forms to make the analysis results intuitive. In order to better link users, we carried out interactive system and visual design, did UI design of web pages and APP, and built interactive prototypes. The main functions of the design are: real-time voice monitoring, intelligent health recognition, historical data analysis and model self-optimization.

How it is different

It excludes the complex emotion and content recognition functions of cat and dog whisperers on the market, focuses on whether the bark is healthy or not, and has low development cost and strong segmentation.

Future plans

The future vision is to be able to expand the variety of animal identification and strengthen the professionalism and accuracy, preferably in cooperation with professional institutions and in combination with smart hardware (such as pet feeders, etc.)

Awards