Automated directory assistance system using word recognition and phoneme processing method

Claims

We claim:

1. A method of providing subscriber telephone numbers to telephone users in an automated fashion comprising the steps of:

(a) connecting a telephone user to an automated directory assistance station upon a user dialing a predetermined number on a telephone;

(b) instructing the user with a stored message upon said connection to respond by speaking a name of a location of a sought subscriber;

(c) encoding a first response from the user into a first digital signal compatible with means for speech recognition;

(d) transmitting said first digital signal to first means for speech recognition by word recognition and second means for speech recognition by phoneme recognition;

(e) decoding first output signals from said first and second means for speech recognition to produce first and second decoded signals and a probability level signal associated with each decoded signal, said probability level signals being indicative of a probability that a respective decoded signal is correct with respect to the first response;

(f) combining said probability level signals associated with said first and second decoded signals according to a predetermined function to derive a plurality of combined probability level signals;

(g) comparing each said combined probability level signal to a predetermined threshold;

(h) selecting one of said first and second decoded signals associated with the highest combined probability level signal to provide a first selected signal;

(i) addressing a database for a location indicated by said first selected signal;

(j) instructing the user with a stored message to respond by speaking the last name of the sought subscriber;

(k) encoding a second response comprising the last name from the user into a second digital signal compatible with said first and second means for speech recognition;

(l) transmitting said second digital signal to the first and second means for speech recognition;

(m) decoding second output signals from said first and second means for speech recognition to produce third and fourth decoded signals and a probability level signal associated with each of said third and fourth decoded signals with respect to said second response;

(n) selecting one of said third and fourth decoded signals having the highest probability level to provide a second selected signal;

(o) searching said database using said second selected signal to obtain a directory number corresponding thereto; and

(p) transmitting a message to the user articulating said directory number.

2. A method according to claim 1 including the steps of:

responding to locating multiple matches in said database as a result of steps (i) or (o) by transmitting to the user a voiced selection of words with a request for an affirmative or negative response.

3. A method according to claim 2 wherein said user is requested to indicate said affirmative response or said negative response by voicing "YES" or "No".

4. A method according to claim 2 wherein said user is requested to indicate said affirmative response or said negative response by striking a DTMF key designated in said request.

5. A method according to claim 1 including the steps of:

responding to locating multiple matches in said database as a result of steps (i) or (o) by requesting the user to speak an additional indicia of the sought subscriber;

encoding a third response from the user into a third digital signal compatible with said first and second means for speech recognition;

transmitting said third digital signal to said first and second means for speech recognition;

decoding and translating a fifth decoded signal responsive to the transmitting of said third digital signal; and

wherein the step (o) further comprises searching said database using said fifth decoded and translated signal.

6. A method of providing subscriber telephone numbers to telephone users in an automated fashion comprising the steps of:

(a) connecting a telephone user dialing a predetermined number on a telephone to an automated directory assistance means;

(b) instructing the user with a stored message upon said connection to respond by speaking an indicia of the location of a sought subscriber;

(c) encoding a first response from the user into a first digital signal compatible with means for speech recognition;

(d) transmitting said first digital signal to first means for speech recognition by phoneme recognition and second means for speech recognition by word recognition;

(e) generating a plurality of candidate words from said second speech recognition means, said candidate words being expressed as encoded signals output by said second speech recognition means related to said first response;

(f) decoding signals output by said first and second speech recognition means responsive to said first digital signal to produce first and a plurality of second decoded signals and a probability level signal associated with each decoded signal, said probability level signals being indicative of a probability that a respective decoded signal is correct with respect to the first response;

(g) comparing said probability level signals from said second speech recognition means to a predetermined threshold level and, if one of said probability level signals is equal to or greater than said threshold level, addressing a database of directory number data for a first location indicated by the decoded signal associated with the one probability level signal greater than or equal to said threshold level;

(h) if all said probability level signals associated with said second plurality of decoded signals are less than said threshold level, combining said probability level signals from said first and second speech recognition means to obtain a plurality of combined probability level signals;

(i) selecting one of said first and second decoded signals having the highest combined probability level signal;

(j) addressing a database of directory number data for a second location indicated by the one selected signal;

(k) instructing the user with a stored message to respond by speaking an indicia of a name of the sought subscriber;

(l) encoding a second response spoken by the user into a second digital signal compatible with said first speech recognition means;

(m) responsive to input of said second digital signal decoding an output of said first speech recognition means to produce a third decoded signal;

(n) using said third decoded signal searching said database for said second location to obtain a directory number corresponding to the second response; and

(o) transmitting a message to the user articulating said directory number.

7. A method according to claim 6 including the steps of:

responding to locating multiple matches in said database as a result of steps (j) or (n) by transmitting to the user an audio selection of words with a request for an affirmative or negative response.

8. A method according to claim 7 wherein said user is requested to indicate said affirmative response or said negative response by voicing "YES" or "NO".

9. A method according to claim 7 wherein said user is requested to indicate said affirmative response or said negative response by striking a key designated in said request.

10. An automated directory assistance system for use in a telecommunications network connected to a plurality of telephone stations comprising:

means for speech recognition, means for voice processing, and computer means for controlling said voice processing means and said speech recognition means;

a database associated with said computer means, said speech recognition means and said voice processing means;

said speech recognition means including means for word recognition and means for phoneme recognition, each encoding voice signals inputted thereto and providing an output candidate signal indicative of a sought word signified by said voice signals and an associated probability level signal indicative of a probability level in the accuracy of the word signified by the output candidate signal, said speech recognition means further including means for calculating a combined probability value signal by combining the probability level signals associated with the output candidate signals from the respective word recognition means and phoneme recognition means to obtain a combined probability level signal for each output candidate signal, the combined probability level signal indicating a recognition accuracy between said output candidate signal of the word recognition means and an associated output candidate signal from said phoneme recognition means;

said database including stored word and phoneme data;

said voice processing means including stored voice messages;

comparator and selector means associated with said word recognition means and said phoneme recognition means for comparing the probability level signals associated with respective the output candidate signals of said word recognition means and said phoneme recognition means responsive to said voice signals and selecting one of said output candidate signals having an associated combined probability level signal with the highest value;

said computer means comprising means for:

(a) causing said voice processing means to transmit, to a user having dialed directory assistance, instructions to speak a word or words indicative of an identity of a subscriber whose directory number is sought;

(b) causing said voice signals resulting from a response from the user to be inputted to said word recognition means and said phoneme recognition means;

(c) causing said word recognition means and said phoneme recognition means to encode said voice signals to produce said respective output candidate signals and said corresponding probability level signals;

(d) to cause said comparator and selector means to compare said probability level signals, the combined probability level signal with the highest value, and access said database using the corresponding selected one output candidate signal to identify a directory number in said database; and

(e) causing said means for voice processing to direct to said user a signal indicative of said directory number.

11. An automated caller assistance system for use in a telecommunications network connected to a plurality of telephone stations comprising:

means for speech recognition, means for voice processing, and computer means for controlling said voice processing means and said speech recognition means;

a database associated with said computer means, said speech recognition means and said voice processing means;

said speech recognition means including means for word recognition and means for phoneme recognition, said word recognition means comprising means for generating a plurality of candidate words for each word spoken by a user, each said candidate word represented by an output candidate signal with an associated candidate probability level signal, said phoneme recognition means comprising means for generating a plurality of phonemes associated with each of said candidate words, each said phoneme being represented by an output phoneme signal and an associated phoneme probability level signal, said speech recognition means further comprising means for combining the candidate probability level signal for each said word candidate with the phoneme probability level of each of said phonemes associated with a respective candidate word according to a predetermined function to obtain a plurality of combined probability level signals for each said candidate word;

said database including stored word and phoneme data;

said voice processing means including stored voice messages;

comparator and selector means, associated with said word recognition means and said phoneme recognition means, for comparing the candidate and phoneme output signals responsive to each word spoken and selecting the candidate word having the corresponding combined probability level signal with the highest value;

said computer comprising means for:

(a) causing said voice processing means to transmit instructions prompting the user to speak a word or words indicative of the nature of the assistance which is sought;

(b) causing said each word resulting from a response from the user to be inputted to said word recognition means and said phoneme recognition means;

(c) causing said word recognition means and said phoneme recognition means to encode said each word to produce the respective candidate and phoneme output signals and candidate and phoneme probability level signals;

(d) causing said comparator and selector means to compare said probability level signals, identify the combined probability level signal with the highest indicia of probability, and associate the candidate word corresponding thereto with said database to identify an assistance associated therewith in said database; and

(e) causing said means for voice processing to direct to said user an audio signal indicative of said assistance.

12. A method for automatically providing subscriber telephone numbers to telephone users over a telephone line, comprising the steps of:

(a) connecting a telephone user dialing a predetermined number to an automated directory assistance station;

(b) transmitting a first response from said user to a speech recognition device comprising means for recognizing a word from among a plurality of words and means for recognizing a phoneme string for association with a respective word;

(c) obtaining a plurality of word candidates related to said first response from the word recognizing means, each of said word candidates having a probability value indicating a probability that said each word candidate is correct with respect to said first response;

(d) selecting one of the word candidates having a highest probability value;

(e) comparing said highest probability value of the selected word candidate to a first threshold value;

(f) if said highest probability value is equal to or greater than said first threshold value, accessing a first database to obtain first information corresponding to said selected word candidate;

(g) transmitting said first information to said user;

(h) if said highest probability value is less than said first threshold value, obtaining at least one phoneme string associated with each said word candidate, each said phoneme string having a probability value indicating a probability that said each phoneme string is correct with respect to said first response;

(i) combining said probability value for each said word candidate and the probability value of the corresponding at least one phoneme string according to a first predetermined function to obtain at least one combined probability value for each said word candidate;

(j) selecting one of said word candidates having the highest combined probability value;

(k) comparing said highest combined probability value to a second threshold value to determine a satisfactory level;

(l) if said highest combined probability value is equal to or greater than said second threshold value, accessing said first database to obtain said first information corresponding to the selected word candidate having the highest combined probability value; and

(m) transmitting said first information to said user.

13. The method of claim 12, further comprising the steps of:

(n) if said highest combined probability value is less than said second threshold value, prompting said user to provide at least a partial spelling of said first response;

(o) transmitting a response including said spelling to means for recognizing individual letters of the alphabet to output word candidate data related to said spelling, said word candidate data having a probability value indicating a probability that said word candidate data is correct with respect to said spelling;

(p) comparing said probability value of said word candidate data output from the recognizing letters means to a third threshold value;

(q) if the probability value associated with said word candidate data from said recognizing letters means is equal to or greater than said third threshold value, accessing said first database to obtain the first information corresponding to said word candidate data.

14. The method of claim 13, further comprising steps of:

(r) if said probability value associated with said word candidate data output from said recognizing letters means is less than said third threshold value, prompting said user to input second alphabetic data via a keypad for transmission to a third database;

(s) accessing said first database to obtain second information corresponding to said second alphabetic data; and

(t) transmitting said second information to said user.

15. The method of claim 12, wherein each said word candidate has a single corresponding phoneme string associated therewith, and said phoneme string is derived based upon phonetic spelling of said associated word candidate.

16. The method of claim 15, wherein each said phoneme string is derived based upon a plurality of pronunciations of said associated word candidate.

17. A method for automatically providing subscriber telephone numbers to telephone users over a telephone line, comprising the steps of:

(a) connecting a telephone user dialing a predetermined number to an automated directory assistance station;

(b) transmitting a first response from said user to a speech recognition device comprising means for recognizing a word from among a plurality of words and means for recognizing a phoneme string for association with a respective word;

(c) obtaining a first plurality of word candidates related to said first response from the word recognizing means and a first plurality of phoneme strings for each said word candidate from the phoneme recognizing means, each of said word candidates and each of said phoneme strings related thereto having probability values indicating a probability that said each word candidate and said each phoneme string are correct with respect to said first response, respectively;

(d) combining said probability value for said each word candidate with said each of said phoneme strings related thereto according to a first predetermined function to obtain a first plurality of combined probability values for said each word candidate;

(e) selecting one of said word candidates having the highest combined probability value;

(f) accessing a first database to obtain first information corresponding to said selected word candidate;

(g) instructing said user to provide a second response related to said first response;

(h) transmitting said second response to said speech recognition device;

(i) accessing said first database to obtain a plurality of second word candidates related to said second response, each of said second word candidates having a plurality of second phoneme strings related thereto, each of said second word candidates and said second phoneme strings having probability values indicating a probability that said each second word candidate and said each second phoneme string are correct with respect to said second response, respectively;

(j) combining said probability value for said each second word candidate with said each of said second phoneme strings related thereto according to said first predetermined function to obtain a second plurality of combined probability values for each said second word candidate;

(k) selecting one of said second word candidates having the highest combined probability of value of said second plurality;

(l) accessing said first database to obtain second information corresponding to the selected one of said second word candidates; and

(m) transmitting said second information to said user.

18. The method of claim 17, wherein said first response is related to a location and said second response is related to the identity of a subscriber.

19. A method for automatically providing directory assistance information including a subscriber telephone number to a telephone user over a telephone line, comprising the steps of:

(a) connecting the telephone user to an automated directory assistance station;

(b) transmitting a first response from said user to means for recognizing speech, said speech recognizing means comprising means for recognizing a word from among a plurality of words;

(c) obtaining a plurality of word candidates related to said first response, each of said word candidates having a probability value indicating a probability that said each word candidate is correct with respect to said first response;

(d) comparing the probability values of said respective word candidates to a first threshold value;

(e) accessing a first database in response to identifying one of the word candidates having a probability value greater than or equal to a first threshold value, and obtaining first directory assistance information corresponding to the one identified word candidate;

(f) transmitting said first information to said user;

and wherein if none of said probability values is greater than or equal to said first threshold:

(g) obtaining at least one phoneme string associated with each said word candidate, said at least one phoneme string having a probability value indicating a probability that said at least one phoneme string is correct with respect to said first response;

(h) combining said probability value for said each word candidate with the probability value of the corresponding at least one phoneme string according to a first predetermined function to obtain at least one combined probability value for said each word candidate;

(i) selecting one of said word candidates having the highest combined probability value;

(j) comparing said highest combined probability value to a second threshold to determine a satisfactory level;

(k) if said highest combined probability value is greater than or equal to said second threshold, accessing said first database to obtain second directory assistance information corresponding to said selected word candidate; and

(l) transmitting said second information to said user.

20. A method for automatically providing subscriber telephone numbers to telephone users over telephone line, comprising the steps of:

(a) connecting a telephone user to an automated directory assistance station;

(b) transmitting a first response from said user to means for recognizing speech, said means for recognizing speech comprising means for recognizing a word from among a plurality of words and means for recognizing a phoneme string for association with a respective word;

(c) obtaining a plurality of word candidates related to said first response, each of said word candidates having a probability value indicating a probability that said each word candidate is correct with respect to said first response;

(d) comparing the probability value of a first of the word candidates with a first threshold;

(e) if said probability value of said first of the word candidates is equal to or greater than said first threshold value, accessing a database to obtain first information corresponding to said first of the word candidates;

(f) instructing the user to provide a second response related to said first response;

(g) transmitting said second response to said means for recognizing speech;

(h) obtaining a plurality of secondary word candidates related to said second response, each of said secondary word candidates having a probability value indicating a probability that said each secondary word candidate is correct with respect to said second response, and obtaining at least one phoneme string associated with said each secondary word candidate, each said phoneme string having a probability value indicating a probability that said each phoneme is correct with respect to said second response;

(i) combining said probability value for said each secondary word candidate with the probability value of the corresponding at least one phoneme string according to a first predetermined function to obtain at least one combined probability value for said each word candidate;

(j) selecting one of said secondary word candidates having a highest combined probability value;

(k) comparing said highest combined probability value to a second threshold to determine a satisfactory level;

(l) if said highest combined probability value is equal to or greater than said second threshold value, accessing said database to obtain second information corresponding to the one secondary word candidate; and

(m) transmitting said second information to said user.

21. An automated directory system for use in a telecommunications network connected to a plurality of telephone stations, comprising:

(a) means for recognizing a word spoken by a user from among a plurality of words, said recognizing means comprising means for generating a plurality of candidate words responsive to the spoken word, and means for assigning a probability value to each of the generated candidate words indicative of a probability that said each candidate word is correct with respect to said spoken word;

(b) means for recognizing phoneme string comprising means for generating a plurality of phoneme strings related to each of said candidate words, and means for generating a probability value for each said phoneme string indicative of a probability that said each phoneme string is correct with respect to said spoken word;

(c) means for combining the probability values of the phoneme strings with probability values of the respective candidate words to obtain a combined probability value for each said candidate word;

(d) comparison means for determining when one of the combined probability values is equal to or greater than a predetermined threshold value;

(e) a database containing information related to the plurality of word candidates; and

(f) means for applying the candidate word corresponding to the one combined probability value to access said first database to obtain information to be sent to said user.

22. The system of claim 21, wherein said means for combining the probability values comprises means for correlating the phoneme strings corresponding to said each candidate word to combine the probability level of said each word candidate with the probability level of each of said corresponding phoneme strings to provide a plurality of combined probability values for said each word candidate.

23. A method for automatically providing directory assistance information including a subscriber telephone number to a telephone user, comprising:

connecting a telephone operated by the telephone user to a directory assistance station;

transmitting a first spoken input from the telephone user to a word recognition device and a phoneme recognition device;

in response to the first spoken input, outputting a plurality of word candidates from the word recognition device and a plurality of phoneme strings for each of the word candidates from the phoneme recognition device, each of the word candidates having a candidate probability value and each of the phoneme strings having a string probability value, each candidate probability value and string probability value indicating a probability that the corresponding word candidate and phoneme string is correct with respect to the first spoken input, respectively;

combining each candidate probability value, in accordance with a predetermined function, with the corresponding plurality of string probability values of the respective plurality of phoneme strings for said each word candidate to obtain a plurality of combined probability values for said each word candidate;

selecting one of the word candidates having a highest combined probability value; and

accessing a database in response to the one word candidate to obtain the directory assistance information corresponding to the first spoken input.

24. The method of claim 23, further comprising determining if the highest combined probability value exceeds a threshold.

25. The method of claim 24, further comprising transmitting the directory assistance information if the highest combined probability value exceeds the threshold.