Analysis of Long Duration Snore Related Signals Based on Formant Features

Snoring is a typical symptom of Obstructive Sleep Apnea-Hypopnea Syndrome (OSAHS) patients, which has motivated numerous researchers focusing on how to diagnose this disorder by acoustic signal analysis methods. As a non-invasive approach, acoustic diagnosis brings a much more comfortable and convenient experience to subjects than the gold standard, polysomnography (PSG). However, there is a more demanding need from doctors to find the variations of the upper airway (UA) during a long duration for OSAHS patients. Formant features have a good performance on indicating the structure variations of UA, which can be regarded as a resonance in the snoring generation model. In this paper, we proposed a long duration analysis method of snore related signals (SRS) method based on formant features. The first three formant frequencies (F1, F2 and F3) are extracted to group the long duration SRS data into different states with the help of K-means method. Each state of SRS data represents a degree of collapse in UA. We found that formant features have distinguished values in different states and the transition possibility calculated by Hidden Markov Models (HMM) between each state is helpful for analysis of long duration SRS data. This method could be effective in analysis of variations in UA for OSAHS patients and establishment of long duration SRS database.