Speech-responsive voice messaging system and method

Claim 2

The method according to claim 1, which includes also Dual-Tone Multi-Frequency (DTMF) signal-responsive voice messaging.

Claim 3

The method according to claim 1 or 2, wherein: step a) comprises the detection of an utterance and the identification of particular command words of phrases to which the utterance may correspond, wherein the steps of detection and/or identification are carried out by a speech recognizer (222), preferably in conjunction with a template library (226) and an autoresponse library (234), and wherein the speech recognizer (222) is preferably controlled by a recognition command generator (142) and a speech and logging supervisor (230).

Claim 7

A speech-responsive voice messaging system (10) comprising a system control unit (100), a disk and voice Input/Output (I/O) control unit (160), a data storage unit (170) upon which a database directory entry and a mailbox for each subscriber reside, at least one digital line card (DLC) (180), a telephony interface controller (TIC) (185) corresponding to each DLC (180), and/or a system manager's terminal (SMT) (250); wherein the elements of the speech responsive voice messaging system (10) are selectively coupled via first control bus (260) and a first data base (262) and wherein each TIC (185) is provided to be coupled to a Central Office (CO) switch (20); wherein the system control unit (100) serves to manage the overall operation of the SRVMS (10), in accordance with system parameter settings received via the SMT (250) and the DLC (180) servers to exchange voice data with the CO switch (20), to process DTMF signals, and to perform speech recognition and logging operations under the direction of the system control unit (100).

Claim 8

The system of claim 7, wherein the system control unit (100) comprises at least one or more of the following components a bus and Direct Memory Access (DMA) controller (110), a processing unit (120), and a memory (130) in which a voice messaging (VM) function library (132), an interpreter (134), a DTMF user interface structure (136), a speech user interface structure (138), and ambiguity resolution user interface structure (140), a recognition command generator (142), a recognition result processor (144), a logging unit (146), a reporting unit (148), a Speech Parameter Block (SPAB) library (150), and a call statistic library (152); wherein the bus and DMA controller (110), the processing unit (120), and each element within the memory (130) is coupled via an internal bus (270), the bus and DMA controller (110) is further coupled to the first data and control busses (260, 262), the SMT (250), and to the reporting system (12); wherein the coupling between the bus and DMA controller (110) and the reporting system (12) preferably includes multiple lines; and/or the DLC (180) comprises a DLC bus controller (190), a DLC processing unit (200), a Coder/Decoder (CODEC) (210), and a DLC memory (220); wherein the DLC memory (220) comprises a speech recognizer (222), a DTMF processor (224), a template library (226), a logging buffer (228), a speech and logging supervisor (230), a phrase expander (232), an auto-response library (234), a Pulse Code Modulation (PCM) data buffer (236), and a signal conditioner (238); wherein each element within the DLC memory (220) is coupled to the DLC bus controller (190) and the DLC processing unit (200) via a second data bus (280), and wherein the DLC bus controller (190) is coupled to the DLC processing unit (200) via a second control bus (282) and to the first data and control busses (260, 262); and wherein the CODEC (210) is coupled to the PCM data buffer (236) and the DLC bus controller (190) for effecting DMA-type operations between the PCM data buffer (236) and the TIC (185).

Claim 9

Method for speech-responsive voice messaging comprising the steps of: a) receiving an incoming call notification from a DLC bus controller (190) in an interpreter (134); b) transferring a reference to an initial Speech Parameter Block (SPAB) (300) to a recognition command generator (142) from the interpreter (134) in response to the call notification; c) selection (500) of the initial SPAB (300) by the recognition command generator (142); d) the recognition command generator (142) retrieves (502) at least one recognizer parameter within the selected SPAB (300); e) the recognition command generator (142) issues (504) a recognition parameter directive to the speech and logging supervisor (230); f) the recognition command generator (142) issues (506) a recognition request; g) a voice messaging function within the VM function library (132) selects a current prompt and issues (508) a prompt notification to a phrase expander (232); h) the phrase expander (232) issues the current prompt to the subscriber via the PCM data buffer (236), the CODEC (210) and the TIC (185); i) the recognition result processor (144) retrieves (510) the candidate result; k) the recognition result processor (144) evaluates (512) the quality of the returned candidate results; preferably the recognition result processor (144) returns a value to the interpreter (134) that indicates the outcome of the evaluation; l) the interpreter (134) determines (514) whether recognition is to be repeated; m) repeating of the steps f) to l) and counting the repetitions, preferably with variations of step g), as long as a predetermined repeat count has not been exceeded; n) if the repeat count has been exceeded, the interpreter (134) transitions to the DTMF user interface and the method is ended; o) if the repeat count has not been exceeded, the interpreter (134) determines (520) whether recognition confirmation is required wherein recognition confirmation is required; if confirmation is not required, the method is continued with step r); if confirmation is required, the method is continued with step p); p) the interpreter (134) selects a position or location within the ambiguity resolution user interface structure (140) and transfers a reference to a confirmation SPAB (300) to the recognition command generator (142); q) the interpreter determines whether the confirmation was successful; if the confirmation was not successful, the method returns to step f); if the confirmation was successful the method continues with step r); r) the interpreter (134) transfers (530) control to a mapping function that maps the best candidate word ID to a digit sequence; s) the interpreter (134) determines (532) whether the mapped digit sequence corresponds to a speech user interface navigation operation; if the mapped digit sequence corresponds to a speech UI navigation operation, the method is continued with step t), if not, with step o); t) the interpreter (134) selects (534) a position or location within the speech user interface; the method is continued with step w); o) the interpreter transfers (536) control to a voice messaging function that corresponds to the mapped digit sequence; v) the interpreter (134) updates (538) a communication count; w) the interpreter (134) determines (540) whether the subscriber's call is complete; if the subscriber's call is complete, the method ends, otherwise, the method is continued with step c).

Claim 10

Method for confirming a speech recognition result comprising the steps of: a) retrieving (700) confirmation SPAB data; b) setting (702) of recognizer parameters; c) determinating a current confirmation prompt and issuing a confirmation prompt notification; d) issuing a recognition request; e) retrieving (706) of recognition results; f) evaluating the retrieved candidate results; g) determinating (712) if the confirmation result was good; if the confirmation result was good, the method is continued with step h), otherwise with step k); h) setting (718) of a successful confirmation status indicator; the method is continued with step m); k) determinating (714) whether reconfirmation is required; if reconfirmation is required, the method preferably continues with step d); if reconfirmation is not required, the method continues with step l); l) setting (716) an unsuccessful confirmation status; m) restoring (720) the recognizer parameters and ending of the method.

Claim 11

Method for evaluating a speech recognition result comprising the steps of: a) determinating (600) whether the candidate result set indicates that an unrecoverable error or a time out condition has occured; if so, the method is continued with step l), if not, the method is continued with step b); b) selecting (620) a (next) candidate result; c) comparing (622) candidate results with current quality thresholds; d) determinating (624) the candidate quality; e) determinating (626) whether another candidate result requires consideration; if so, the method continues with step b), if not, the method continues with step f); f) determinating (630) whether at least one candidate result has been designated as good; if so, the method continues with step k), if not, the method continues with step g); g) determinating (640) whether a questionable candidate result is present; if so, the method continues with step h), if not, the method continues with step l); h) setting (642) a confirmation required status indicator; i) returning one of the good candidates; ending of the method; k) setting (638) a good result status; the method is continued with step i); l) setting (602) a bad result status; m) incrementing (604) a repeat count; n) determinating (606) whether a predetermined limit is exceeded; if so, the method is continued with step p); if not, the method is continued with step o); o) setting (608) of a repeat status; and ending of the method; p) setting (610) a limit exceeded status; and ending of the method; wherein preferably in step f), "k" is replaced by "q" and in step k), "i" is replaced by "t" with the additional steps of: q) determinating (632) whether there are multiple good candidates; if so, the method is continued with step r), if not, the method is continued with step k); r) examinating (634) of the score differences between each good candidate result; s) determinating (636) whether a difference threshold is exceeded; if so, the method is continued with step k), if the difference threshold is not exceeded, the method is continued with step h); t) returning (644) of the best candidate; and ending of the method.

Claim 12

The method according to claim 10, wherein step f) is performed according to the method of claim 11.

Claim 14

A method of utterance logging comprising the steps of: a) examining (800) logging parameters to determine whether utterance logging is required during a current call; if not the method ends; if utterance logging is required, the method is continued with step b); b) establishing (802) of logging conditions; c) determinating (804) whether the next recognition event is to he logged; if not, the method is continued with step k), if the next recognition event is to be logged, the method continues with step d); d) setting (806) a set of at least one sampling parameters; e) determinating (808) whether the quality evaluation for the current candidate results has been completed; if not, the method preferably remains at step e), if the quality evaluation has been completed the method continues with step f); f) determinating (810) whether any criteria specified in the logging parameters are matched; if not, the method is continued with step k); if any criteria specified in the logging parameters are matched, the method is continued with step g); g) saving (812) of the utterance; h) examining of the current logging parameters and determination whether the saved utterance should be retained for later use; if the utterance should not be retained, the method continues with step k), otherwise with step i); i) updating (816) of the logging buffer storage location referenced to a next available location; k) determinating (818) whether the current call is complete; if not, the method preferably continues with step c); if the call is complete the method continues with step l); l) generating (820) call header information; m) transferring (822) the logging data to a data storage device; end of the method; wherein: preferably the method is performed by means of a logging unit (146) and a speech and logging supervisor (230) which control the selective logging of utterances in accordance with the logging parameters specified in each of a group of SPABs (300); preferably the logging unit (146) operates transparently during a call, monitoring the operation of the interpreter (134), the recognition command generator (142) and the recognition result processor (144); the call header information comprises preferably a mailbox ID, and possibly a reference to a Customer Data Record (CDR), wherein the customer data records are preferably generated by the method of claim 13.

Claim 15

The method according to any of claims 1 to 6 or 9, comprising the method of claim 11 and/or the method of claim 10 or 12 and/or the method of claim 14 and/or the method of claim 13.