AI Thermometer

This app listens when you pour hot water into a cup and analyzes the pouring noise to estimate the water temperature.

The accuracy reaches ±4 degrees when tested under optimal conditions but is generally lower. The accuracy is higher at high temperatures and it can be useful for e.g. brewing different teas at their correct temperatures but it is not intended to be used for temperatures less than 50°C (122°F).

The app visualizes a probability distribution and gives an 80% confidence interval for the temperature that depends on the vessel, pouring technique, and background noise. The probability distribution has been verified on a wide range of setups, but still assumes a prior distribution of pouring setups.

How does it work?

The properties of water and air vary with temperature. This results in changes in how noise is generated from pouring water. This app uses a pre-trained machine learning model to analyze pouring noises to get an estimate of the temperature. The longer time you pour, the more noise it receives, and it can give better estimates.