A new criterion for speech frame classification based on Fuzzy Logic is presented. An algorithm which correctly discriminates between sounds allows a better trade-off between quality and bit rate in the recent generation of vocoders. Extracting from speech waveforms a set of significant parameters for voiced/unvoiced classification, we trained a Fuzzy Voicing Detector (FVD) system, obtaining a fuzzy knowledge base of only 5 rules. The new voicing system has certain advantages over the main Voicing Determination Algorithms (VDAs). With a notoriously low level of complexity, on account of its fuzzy nature, the FVD is extremely robust and maintains good performance even in the presence of background noise.
Bibliographic reference. Beritelli, Francesco / Casale, Salvatore / Russo, Marco (1995): "A voiced/unvoiced speech discrimination technique based on fuzzy logic", In EUROSPEECH-1995, 389-392.