Prosodic Speech Restoration Device: Glottal Excitation Restoration Using a Multi-resolution Approach

We describe a novel device for the restoration of authentic characteristics in pathological speech uttered by subjects with laryngeal disorders. The original speech signal is acquired and analyzed by the device and a speech signal with improved, healthy-like features is reconstructed. A concatenation of randomly chosen healthy reference patterns replaces the pathological excitation. In order to restore authentic features, intervals between subsequent reference patterns are obtained through a multi-resolution approach. Short-term pitch variability is reproduced through a statistical variation model. For middle-term pitch variability the correlation between pitch and signal envelope at the middle-term time scale is exploited. Long-term variability is obtained through adaptive wavetable oscillators; a novel, reliable and computationally efficient pitch estimation method. Two authentic features were used to assess performance, namely breathiness and prosody. Preliminary results indicate that breathiness of the restored signal is clearly reduced and prosody related features are improved.