Can You Hear Me Now? Solving the Headphone Problem

Abstract

This project will develop a product that will be embedded into headphones to prevent the wearer from missing essential communication, either from conversation or emergency sirens (fire alarms, tornado sirens, emergency vehicle sounds, etc.). It will monitor ambient sound and determine whether or not music should be interrupted. A microphone will pick up sound from the environment, then analysis software will process it to determine the next step. If the sound is determined to be either close range human speech or an emergency siren, the code will signal the music to cease. If neither of these is recognized, the music will continue. The bulk of the project lies in developing software that can differentiate human speech from other ambient sounds. Thus far, we are writing code and figuring out what distinguishes human speech from other ambient noise. This includes spectrogram analysis of various sound recordings. From our research, we know that we must analyze each sound to approximately 1/100th of a second (this is defined to be a "feature"). Each feature is assigned a probability that it appears as a portion of a spoken word. We cross-reference each probability with 1000's of feature charts of recognizable human speech patterns to determine if the sound is from a human voice. Based on this probability, we decide if the word is human speech. The code we create to do this will be transferred to a pic microcontroller, which interfaces with the microphone and the music player, allowing it to quickly respond to human speech or emergency signals.