Characteristics of spoken man-machine communication
Automatic speech recognition

Automatic speech processing covers a vast range of often complementary activities, which can be classified into five main themes:

• speech coding and compression, involved in speech transmission and reproduction;

• speech synthesis, consisting in producing a speech signal from a dictionary of phonetic elements and/or rules ;

• speech recognition, in which a machine interprets words or phrases spoken by a human speaker in order to perform a specific action;

• speaker recognition and verification, a dual problem of the previous one in which we seek to authenticate a person by his voice;

• identification of the language in which a speaker expresses himself.

Significant progress has been made in all these fields in recent years, and numerous industrial applications exist. We are interested in the problem...