Audio dialogue system and voice browsing method

• US 20070136067A1
• Filed: 11/09/2004
• Published: 06/14/2007
• Est. Priority Date: 11/10/2003
• Status: Abandoned Application

• Current Assignee: Koninklijke Philips Electronics N.V. (Koninklijke Philips N.V.)
• Original Assignee: Koninklijke Philips Electronics N.V. (Koninklijke Philips N.V.)
• Inventors: Scholl, Holger
• Application Number:
  US10/578,375
• Publication Number:
  US 20070136067A1
• Time in Patent Office: Days
• US Class Current: 704/270
• CPC Class Codes: G10L 15/26 Speech to text systems G10L15/08 takes precedence
• CPC Class Codes: H04M 2201/60 Medium conversion
• CPC Class Codes: H04M 3/4938 comprising a voice browser which renders and interprets, e.g. VoiceXML

Abstract

An audio dialog system and a voice browsing method are described. An audio input unit (12) acquires an audio input signal. Speech recognition means (20) convert the audio input signal into text input data (21). Content data (D1) comprises text content and at least one reference (LN1). The reference comprises a reference aim and an activation phrase. Browsing means (22) process the content data (D1), controlling speech synthesis means (26) to output the text content. The browsing means (22) compare acquired input text data (21) to the activation phrase (28). If the input text data (21) is not identical to activation phrase (28), a match is still indicated if the input text data and the activation phrase have a similar meaning. In case of a match, content data corresponding to the reference aim is accessed.

Claims

1. Audio dialogue system, comprising

an audio input unit (12) for inputting an audio input signal,

speech recognition means (20) associated with said audio input unit (12) for converting said audio input signal into a text input data (21),

an audio output unit (12) for outputting an audio output signal, and speech synthesis means (26) associated with an output unit (12) for converting text output data (24) into said audio output signal,

browsing means (22) for processing content data (D1), said content data (D1) comprising text content and at least one reference (Ln1, Ln2), said reference comprising a reference aim and activation information, said activation information comprising one or more activation phrases (28),

said browsing means (22) being configured to control said speech synthesis means (26) to output said text content,

said browsing means being further configured to compare said input text data (21) to said activation phrase (28), and in case of a match, for accessing content data (D2) corresponding to said reference aim,

where in case that said text input data (21) is not identical to said activation phrase (28), said browsing means (22) find a match, if said input text data (21) has a meaning similar to said activation phrase (28).

2. System according to claim 1, said system further comprising

dictionary means (30) for storing, for a plurality of search words (32a), connected words (32b, 32c, 32d) with a meaning connected to the meaning of said search words (32a),

where said browsing means (22) are configured to retrieve connected words (32b, 32c, 32d) for words comprised in said input text data (21) and/or for words comprised in said activation phrase (28),

and use said connected words (32b, 32c, 32d) for said comparison.

3. System according to claim 2, where

said dictionary means (30) comprise for at least some of said search words (32a),

connected words (32b, 32c, 32d) which fall into one or more of the categories out of the group consisting of:

synonyms, hyponyms, hypernyms, holonyms, meronyms.

4. System according to claim 1, where

said browsing means (22) are configured to establish a co-occurrence matrix giving for a plurality of terms and for a plurality of activation phrases the number of occurrences of said terms in said phrases,

perform a singular value decomposition of said co-occurrence matrix to calculate a semantic space,

and determine a similarity by representing said input text data (21) and said activation phrase (28) as vectors in said semantic space, and calculating a measure for the angle between these vectors.

5. System according to claim 1, where

said browsing means (22) are configured to determine a word frequency for a plurality of words in all activation phrases of all links in said content data,

and determine a similarity by finding common words in said input text data (21) and said activation phrase (28).

6. System according to claim 1, where

said browsing means (22) are configured to determine a word sequence frequency for a plurality of word sequences of all activation phrases (28) of all of said links in said content data,

and determine a similarity by processing word sequences of said input text data (21).

7. System according to claim 1, where

for each of said links a language model is trained, said language model comprising word sequence frequencies,

and said input text data (21) is compared to each of said language models by determining a score indicating an agreement of said input text data (21) with said model,

and said similar meaning is determined according to said score.

8. Voice browsing method, comprising:

processing content data (D1), said content data (D1) comprising text content and at least one reference (LN1), said reference comprising a reference aim and activation information, said activation information comprising one or more activation phrase (28),

converting said text content to an audio output signal using speech synthesis, and outputting said audio output signal,

acquiring an audio input signal, and using speech recognition to convert said audio input signal to text input data (21),

comparing said text input data (21) to said activation phrase (28) and in case that said text input data is not identical to said activation phrase (28), indicating a match if said input text data (21) has a meaning similar to said activation phrase (28), and in case of a match accessing content data (D2) corresponding to said reference aim.