Techniques for disambiguating speech input using multimodal interfaces

• Main Authors: SIBAL, SANDEEP, VAIDYA, SHIRISH, DOMINACH, RICHARD, ISUKAPALLI, SASTRY
• Format: Patent
• Language: English; French; German
• Published: 21.01.2009
• Subjects: ACOUSTICS; MUSICAL INSTRUMENTS; PHYSICS; SPEECH ANALYSIS OR SYNTHESIS; SPEECH OR AUDIO CODING OR DECODING; SPEECH OR VOICE PROCESSING; SPEECH RECOGNITION

A system for disambiguating speech input. The system comprises a speech recognition component (110) for receiving recorded audio or speech input (104) and for generating one or more tokens corresponding to said input and a confidence value for each of said one or more tokens, the confidence value being indicative of a likelihood that said token correctly represents the respective input. The system also comprises a selection component (116) for identifying, according to a selection algorithm, which of any two or more tokens generated for said input are to be presented to a user (108) as alternatives (120); one or more disambiguation components (118,124) for performing an interaction with the user, in which the alternatives are presented to the user and the user's selection (122) is received; and an output interface (126) for presenting the user's selection as an input to an application (106). The system is characterised in that said interaction with the user uses a multimodal interface and said alternatives are presented to the user as a multimodal output and the user's selection is received as a multimodal input.