Skip to Main Content
Helium speech is produced by speaking in a high-pressure helium-oxygen environment. A variety of acoustical and psychological effects render unprocessed helium speech virtually unintelligible. In this paper, a new system for enhancing helium speech based on the short-time Fourier transform (STFT) is described. The proposed system is able to correct many of the helium speech phenomena, does not require either pitch detection or a voiced/unvoiced decision, and is able to incorporate "spectral substraction" noise reduction algorithms in a natural way. Alternative methods for estimating the spectral envelope from the STFT and correcting its distortion are described, as well as a technique for reducing the signal sampling rate at the earliest possible stage.