Speech Reconstruction in Post-Laryngectomised Patients by Formant Manipulation and Pitch Profile Generation

rehabilitation of the ability to speak in a natural sounding voice, for patients who suffer larynx and voice box deficiencies, has long been a dream for both patients and researchers working in this field. Removal of, or damage to, the voice box in a surgical operation such as laryngectomy, affects the pitch generation mechanism of the human voice production system. Such patients speech thus becomes hoarse, whisper like and sometimes not easily perceptible. This speech is obviously different to that from normal speakers, and will have lost many of the distinctive characteristics of the original speech. However, these patients typically retain the ability to whisper in a similar way to normal speakers. This paper aims to present an engineering approach to providing laryngectomy patients the capacity to regain their ability to speak with a more natural voice, and as a side effect, to allow them to conveniently use a mobile phone for communications. The method uses auditory information only, allied with analysis, formant insertion and novel methods for spectrum enhancement and formant smoothing within the reconstruction process. In effect, natural sounding speech is obtained from their spoken whisper-speech, without recourse to surgery. The method builds upon our previously published works using an analysis-by-synthesis approach for voice reconstruction with a modified CELP codec.