8th European Conference on Speech Communication and Technology

Geneva, Switzerland
September 1-4, 2003


Very-Low-Rate Speech Compression by Indexation of Polyphones

Charles du Jeu, Maurice Charbit, Gérard Chollet


Speech coding by indexation has proven to lower the rate of speech compression drastically. Based on the Automatic Language Independent Speech Processing (A.L.I.S.P) approach that automatically segments the speech signal ([1]), we studied the possibility of optimising this rate as well as the quality of re-synthesised signal, by using the text information corresponding to the speech signal, and by implementing a new segmentation method. This led to the speech alignment with its phonetic transcription and the use of polyphones, to finally increase output speech quality while keeping a bitrate between 400bits/s and 600bits/s. Typically, this can be used to store recorded alpha-numeric books for blind people, or compressing recorded courses for e-learning. Cell phone applications could also be considered.

Full Paper

Bibliographic reference.  Jeu, Charles du / Charbit, Maurice / Chollet, Gérard (2003): "Very-low-rate speech compression by indexation of polyphones", In EUROSPEECH-2003, 1085-1088.