Publication Details
Advances in very low bit-rate speech coding using recognition and synthesis techniques
Capman Francois (THALES-COM)
Černocký Jan, prof. Dr. Ing. (DCGM FIT BUT)
El Chami Fadi (ULIBA)
Charbit Maurice (GET/ENST)
Chollet Gerard, Dr. (GET/ENST)
Petrovska-Delacretaz Dijana, Dr. (unifr)
speech coding, very low bit-rate, data-driven units, ALISP
Many current systems for automatic speech processing rely on sub-word units defined using phonetic knowledge. Our paper presents an alternative to this approach -- determination of speech units using {ALISP} (Automatic Language Independent Speech Processing) techniques. Such units were experimentally tested in a very low bit rate phonetic vocoder, where mean bit rates of hundreds bps for unit encoding were achieved. Improvements of the proposed coder and some links to ``classical'' approaches of speech synthesis are discussed. Based on the results of comparison of an ALISP segmentation with a phonetic alignment, we comment on the potential use of automatically derived units in speech recognition, speaker verification and language identification.
@ARTICLE{FITPUB7024, author = "Genevieve Baudoin and Francois Capman and Jan \v{C}ernock\'{y} and Fadi Chami El and Maurice Charbit and Gerard Chollet and Dijana Petrovska-Delacretaz", title = "Advances in very low bit-rate speech coding using recognition and synthesis techniques", pages = "269--276", booktitle = "Proc. 5th International Conference Text, Speech and Dialogue, TSD2002", journal = "Lecture Notes in Computer Science", volume = 2002, number = 2448, year = 2002, publisher = "Springer Verlag", ISBN = "3-540-44129-8", ISSN = "0302-9743", language = "english", url = "https://www.fit.vut.cz/research/publication/7024" }