Hierarchical Recognition of French Vowels by Expert System Iroise-Serac


  • Anne Bonneau
  • Mario Rossi
  • Guy Mercier


We are presenting here an implementation of a French vowel recognition program under IROISE, an expert system for acoustic-phonetic decoding, used in CNET. The rules for recognition are based on polycontextual non-formantic cues; the data are output from a 14-channel vocoder. The algorithm is represented by a binary tree with 37 hierarchized cues. A rule under IROISE represents a branch of the tree. The first one follows the branch defined only by positive cues; the second one puts the list of the first rule in its contextual part by eliminating the last cue. If the rule is applied we know that this cue is negative, because the preceeding rule was not set off, and we modify the cue's polarity. With this method, only the cues tested in the recognition phase will have the value "false". Under IROISE, all cues are systematically tested even if they are not all used in any particular execution of the program. Then we call the algorithm in which every rule represents a branch. We furnish the recognition results using this program on an initial corpus of 330 words pronounced by five male speakers and the results using rules under IROISE on digits pronounced by other speakers.

Additional Files



How to Cite

Bonneau A, Rossi M, Mercier G. Hierarchical Recognition of French Vowels by Expert System Iroise-Serac. Canadian Acoustics [Internet]. 2022 Dec. 3 [cited 2024 May 23];14(3 bis):20-1. Available from: https://jcaa.caa-aca.ca/index.php/jcaa/article/view/3503



Proceedings of the Acoustics Week in Canada