Method and apparatus for extracting voiced telephone numbers and email addresses from voice mail messages

The present invention relates in general to voice mail systems and more specifically to a method and apparatus for identifying and extracting voiced telephone numbers and email addresses from recorded voice mail messages.

A common complaint among users of voice mail systems relates to difficulties in extracting and transcribing important information from voice mail messages for later use. As an example, it is common for a caller whose call is answered by voice mail to leave a voice mail message that includes a telephone number or email address at which the caller may be contacted. Unless the caller enunciates the telephone number or email address very slowly, the voice mail recipient may be forced to replay all or a portion of the voice mail message in order to transcribe the telephone number or email address for later use. Most callers find very slow enunciation to be awkward and unnatural, with the result that the majority of such voice mail messages must be replayed once or even several times in order for the voice mail recipient to transcribe the information. Furthermore, the process for replaying voice mail messages is often complex (e.g. requiring memorization of keys on a telephone keypad which have been assigned for the replay function). The process can also be tedious (e.g. some systems require that the entire message be replayed in order to extract only the portion relating to the telephone number or email address).

According to the present invention, a system is provided for automatically identifying enunciation of telephone numbers and email addresses in voice mail messages for presentation to a user via either a voice or text, without the user being required to replay the message. Furthermore, according to the invention the extracted telephone number or email addresses may be forwarded to an application for further processing (e.g. to dial the extracted number, add the email address and/or telephone number to a contacts database, etc.)

According to another aspect of the invention, a user may specify text strings to be searched (other than telephone numbers and email addresses). Thus, a user can specify certain words such as “extension”, “email”, etc., in order to qualify the extracted number or email address information, or words such as “urgently” or “immediately” to identify important voice mail messages.

A description of a preferred embodiment of the present invention is provided herein below with reference to the accompanying drawings in which:

FIG. 1 is a schematic representation of a PBX voice mail system incorporating telephone number extraction in accordance with a preferred embodiment of the present invention;

FIG. 2 is a graphical user interface showing an example of the manner in which extracted information is presented to the user according to the invention;

FIG. 3 is a graphical user interface showing an example of user selection of an action to be performed using the extracted information shown in FIG. 2; and

FIG. 4 is a graphical user interface showing an initialization screen for user programming of information to be extracted from voice mail messages.

With reference to FIG. 1, a server based PBX is shown (i.e. PBX) comprising a server 1 , to which a plurality of communication endpoints are connected such as telephone 3 and PC 5 . The telephone 3 can be a digital set, POTS set, a PC phone or LAN/IP phone. For ease of illustration and description, only two endpoints are depicted in FIG. 1 . In practice, a multiplicity of endpoint devices is provided. The server 1 establishes communication between the endpoints under control of call processing software 7 , in a well-known manner.

According to the invention, a voice number extraction voicemail application 9 uses an audio file 11 , knowledge database 12 , and sound database 13 , the operations of which are set forth in greater detail below. The telephone 3 includes a plurality of feature keys for executing commands in connection with the voice number extraction application 9 . Specifically, a “Listen to Message” key 15 is provided which, when pressed by a telephone user causes DTMF digits (or other signaling) to be sent by the telephone 3 to the call processing software 7 . The call processing software 7 then activates the application 9 to retrieve voice mail messages from audio file 11 .

In response to pressing a “Find Number” key 17 while listening to a voice mail message, DTMF digits (or other signaling) are sent by the telephone 3 to the call processing software 7 . A keypad access code, or soft key on telephone 3 , or a GUI button or menu item on PC 5 may replace the key 17 . The call processing software 7 then causes the application 9 to launch a voice recognition application (e.g. Speak@Ease by Mitel Corporation) with instructions to search for a number in the retrieved voice mail message, as described in greater detail below. When the voice recognition application finds a number, it is presented to the user at telephone 3 or PC 5 via voice or textual display (provided the phone 3 supports textual display). It is contemplated that the extracted phone number may be validated against the phone number extracted from Caller ID data that may be stored in the voice message. Upon extraction of a number, the application 9 generates a user prompt for further instructions. The user prompt can be either voice or text. In response to the prompt, the user can select from one of several feature keys (or access codes, or soft keys or GUI buttons or menu items) identified in FIG. 1 as “Accept Number” key 19 , “Continue Search” key 21 , or “Abandon Search” key 23 .

On selecting the “Accept Number” key 19 while listening to a voice mail message, DTMF digits (or other signaling) are sent by the telephone 3 to the call processing software 7 . The software 7 then causes the application 9 to accept a number extracted from a voice mail message as a result of the prior activation of the “Find Number” key 17 . The extracted number (or text for e-mail, etc.) may then be used to trigger further actions, as discussed in greater detail below with reference to FIGS. 2 and 3.

In response to pressing the “Continue Search” key 21 while listening or after having listened to a voice mail message, DTMF digits (or other signaling) are sent by the telephone 3 to the call processing software 7 . The software 7 then causes the application 9 to continue searching for further voiced numbers in a voice mail message after an initial voiced number or numbers have been extracted.

Upon pressing the “Abandon Search” key 23 while listening or after having listened to a voice mail message, DTMF digits (or other signaling) are sent by the telephone 3 to the call processing software 7 which, in response causes the application 9 to abandon any voiced number search underway.

In response to pressing the “Set Phrases” key 25 while listening or after having listened to a voice mail message, DTMF digits (or other signaling) are sent by the telephone 3 to the call processing software 7 . The software 7 then causes the application 9 to display a graphical user interface (FIG. 4) for assignment of voiced phrases to specific pattern items (e.g., extension, local phone number, email address, etc.).

Upon pressing the “See Full List” key 27 , while listening or after having listened to a voice mail message, DTMF digits (or other signaling) are sent by the telephone 3 to the call processing software 7 . The software 7 then causes the application 9 to display a graphical user interface (FIG. 3) for listing all of the assigned voiced phrases and associated items.

As indicated above, the present invention utilizes a voice recognition application (e.g. Speak@Ease by Mitel Corporation) with instructions to search for a number in the retrieved voice mail message. It is well known that voice signals can be divided up into sections, cslu.cse.ogi.edu/tutordemos/nnet_recog/recog, Jul. 6, 1999, for analysis. Voice is usually stored in digital format, or can be converted from analog to digital format for storage in audio file 11 . Thus, when the “Find Number” key 17 is pressed, the call processor 7 launches the voice number extraction application 9 which analyzes the voice message by dividing the message into detected speech patterns. These speech patterns are compared to data in knowledge database 12 using grammar representations (i.e. so-called “hot phrases” representing) programmed by the user, as discussed in greater detail below. The sound database 13 contains sound samples and algorithms, which are used by the application 9 to evaluate speech patterns, as discussed in greater detail below.

The grammar representations (FIG. 4) each have associated tracking pointers (i.e. pointers to symbols in a phrase) at the beginning of each sound representation. For example, in FIG. 4 the phrase #### has a tracking pointer pointing to the first ‘#’ in the phrase, where ‘#’ represents a number wildcard which has been specified in a user programmed grammar representation. If a number is found in the voice message (e.g. the voiced number ‘1’), then the tracking pointer points to the next ‘#’ in the phrase. This means that the next sound heard in the voice mail must be a number in order for the application 9 to detect a match. The tracking pointer defaults to the beginning of the phrase, but can be positioned at a specified point in a phrase identified by a “hot symbol” such as “@” shown in FIG. 4, in which case the tracking pointer points to points adjacent the “hot symbol” on either side thereof. The drop-down actions shown in FIG. 4 contain pre-defined pattern types. For example, when a pattern of nnnn is entered, then the pattern type is defaulted to an extension. However, the user may change the default to something else.

If two or more phrases are extracted, the first phrase is displayed and if the user chooses the ‘continue search’ key 21 , the application 9 displays the second extracted phrase.

As shown in FIGS. 2 and 3, extracted phrases can be selected by the user to trigger further actions in other applications and use the ‘extracted’ information. In that regard, the drop-down actions displayed are context sensitive. For example, the extracted phrase “external” has as valid actions: ‘Call Number’, ‘Phone Book’, etc. The extracted phrase “e-mail” has ‘Reply’, ‘Block e-mail Address’, etc. as actions. The user may add other actions.

In operation, the voice number extraction application 9 loads consecutive speech patterns into positions of a memory (e.g. a FIFO), up to the maximum number of symbols specified by the user in the grammar representation (FIG. 4 ). For each group of stored symbols, the individual stored sounds, symbols, numbers, letters or words are compared with any user identified “hot phrases”, which are groupings of “hot symbols” and wildcards. The “hot symbols” within each hot phrase are compared to the counterpart symbols in the stored sound block. If all of the hot symbols match, then all of the sound blocks are analyzed to determine what the message says. The phrase is then bookmarked for listening.

For example, assume the grammars specified by the user are: phrase 1=***@**.* and phrase 2=##@**.on.ca. The first phrase has eight blocks of sound required (six *'s and two symbols (‘@’ and ‘.’)), whereas the second phrase has eleven blocks of sound required (two *'s, two #'s (any 2 numbers), and seven symbols (‘@’, two ‘.’, ‘o’, ‘n’, ‘c’, and ‘a’)). Therefore, the maximum number of sound blocks needed is eleven blocks. Consequently, eleven sound blocks are saved sequentially as each sound block is being analyzed, as follows:

On the eighth block of sound from the voice mail message phrase 1 stored in sound data file 13 is compared with the sound samples in the block range 1-8. Sound block 4 is compared with ‘@’ (pronounced ‘at’) and sound block 7 is compared with ‘•’ (pronounced ‘dot’). Since there is no match in the illustrated example, the next sound bite is saved to block 9 , as follows

The application 9 then compares ‘@’ with sound block 5 and ‘•’ with sound block 8 . Again, in the illustrated example there is no match.

After several block comparisons, the stored phrase segments become as follows:

The application 9 compares ‘@’ with sound block 7 and ‘.’ with sound block 10 . Again, in the illustrated example there is no match. At this stage, a sufficient number of sound blocks have been stored that phrase 2 may begin to be compared. In particular, phrase 2 compares:

Block 1 to see if it is a number.

Block 2 to see if it is a number.

Block 3 with ‘@’ (pronounced ‘at’).

Block 6 with ‘•’ (pronounced ‘dot’).

Block 7 with ‘o’.

Block 8 with ‘n’.

Block 9 with ‘.’ (pronounced ‘dot’).

Block 10 with ‘c’.

Block 10 with ‘a’.

NOTE: in practice, for efficiency, the comparison with other blocks is continued only if a match is found for Block 1 .

The sound block list is sequentially pushed through the FIFO (first in, first out) stack. Thus, the phrase ‘Hi’ disappears from the stack as the list is pushed to the left and the next sound element from the voice message is queued (i.e. the phrase ‘at’ in block 11 ).

After additional comparisons and shifts, the phrases stored in the FIFO reach a position at which there is a match, as indicated above. The phase ‘reach’ used to be in sound block 9 but has now been queued to sound block 1 .

Phrase 1 matches because:

Block 7 is an ‘@’. pronounced ‘at’)

Block 10 is a ‘.’. (pronounced ‘dot’)

For a user listening to a voice mail message, the pseudo-code for each of the call processing block 7 and voice number extraction block 9 is as follows:

It will be appreciated that, although embodiments of the invention have been described and illustrated in detail, various changes and modifications may be made. All such changes and modifications may be made without departing from the sphere and scope of the invention as defined by the claims appended herein.