Esophageal voices: glottal flow restoration

One of the main problems when working with esophageal voices is the uselessness of traditional speech processing techniques. Concretely, when trying to delete noise from the speech signal, spectral subtraction techniques are not valid because the breathing noise which appears between syllables does not follow the same patterns as the one which appears during the speech. A new technique is needed that allows us to eliminate the breathing noise, which seriously affects later processing of this speech. The aim of this paper is to apply a new technique, which is based on a spectral match with esophageal noise patterns, to the treatment of noisy intervals between speech periods and the deleting of the glottal air flow characteristic spectral components.