Astra Blaster

home *** CD-ROM | disk | FTP | other *** search

/ Astra Blaster / Astra.cdr / voice / hearsay2.txt < prev next >

Wrap

Text File | 1992-04-01 | 26.1 KB | 548 lines

2.00 HOW TO CREATE HEARSAY GOLD 2.01 HOW TO SELECT WORDS IN YOUR VOCABULARY The single most important factor in optimizing the performance of the Hearsay Gold voice recognizer is the selection of vocabulary names. This section provides guidelines that will lead to consistently higher recognition performance in your applications. AVOID SIMILIAR SOUNDING NAMES Remember to consider how the words sound, not how they are spelled. Avoid reliance on final sounds to distinguish words. In normal pronunciation they are often left off words. Consider the words "Eight" and "Aid". When they are carefully pronounced they are quite different from each other, but when said quickly or casually they are very similiar. Other examples would be "Harp", "Heart", and "Hard". AVOID VERY SHORT WORDS When listening for words, Hearsay Gold ignores a very short burst of sound as background noise. If a name is very short, it may sometimes be ignored, such as the word "Up". If you choose "Page up" instead of "Up" you will achieve higher performance. In general, the more syllables in a word, the high the recognizer performance will be. NOTE: Many other two letter words do not have this problem (eg. to, my, etc.). BE CAREFUL OF UNSTRESSED SYLLABLES If a word begins or ends with an unstressed syllable, sometimes the Hearsay Gold may not hear the syllable. The first syllable of the word "About", is unstressed and usually low in volume. In the presence of background noise, Hearsay Gold may hear the word as "Bout". If the word "Out" is present in the same vocabulary, the system may decide that what it hears is closer to "Out" than it is to "About". One strategy for avoiding this difficulty is to put an additional word in front of the problem word such as "Tell about" instead of "About". 2.02 HOW TO DEFINE HEARSAY GOLD VOCABULARIES A Hearsay Gold Vocabulary is a list of words or voice commands that execute keyboard commands. Each word consists of the following: * A name * A template or the users stored voice recording * Keystrokes that are automatically entered into the PC when the voice command is recognized Vocabularies are saved in files known as Vocabulary files. Vocabulary files have an extension of .VOC and contain all of the information associated with a particular vocabulary. Vocabulary files are created and edited using the Hearsay Gold Editor program, which is explained briefly in this section and in depth in section 5.00. A Hearsay Gold vocabulary can contain up to 64 seperate voice commands. If more than 64 commands are needed, you may define multiple vocabularies, and load them into the library. Library vocabularies can be used to "overlay" the current vocabulary automatically on a word by word basis if necessary. This allows a given vocabulary to be changed to suit the changing syntax of the application. For information on this feature and how to use it, refer to section 2.05 HOW TO LINK VOCABULARIES. A Vocabulary also contains other information general to the particular set of commands. Examples of this type of information would be a mouse sensitivity or scale factor parameters is described in section 5.02. In order to create a vocabulary the appropriate recognition driver must be loaded first. The drivers are incorporated in the two startup batch files created in SETUP called XXXE and XXX. XXXE: (Where XXX are your initials) loads a driver that includes an on line vocabulary edit capability described in section 5.00. This driver allows vocabularies to be created or edited without exiting from application programs. When creating a new vocabulary where changes are inevitable it is wise to use this driver instead of XXX. XXX: Loads a driver that does not include an on line editor. In order to edit using XXX you must exit your program then load "RE" which is described below. NOTE: XXXE takes up considerably more memory than the XXX driver (About 40K Bytes more. It should be only used for development when the memory is available. After developing a vocabulary it should be saved then reloaded using XXX. After loading the appropriate driver memory space for the library must be allocated in the computer. This is accomplished by use of the MKLIBXX statement (Where XX is the amount of memory in K bytes). 25K would be enough space required for the first 64 word vocabulary and an additional 12K for each additional 64 word vocabulary. NOTE: After final creation of a vocabulary the allocation of memory will be automatically specified by use of the loading command REDT (See section 3.00 and the MKLIB command is not necessary. is used only for development of new vocabularies or when adding to existing vocabularies. TO DEFINE A HEARSAY GOLD VOCABULARY: In order to define a vocabulary the Hearsay Editor must be loaded. The Hearsay Editor is a utility program which is used to define, edit, train, load, and save Hearsay Gold vocabularies. It can also be used to erase the Hearsay Gold Library. The Hearsay Editor program is included on the Hearsay Gold system disk under the name "REDT.EXE" and "RE.EXE". The two most important areas of the Hearsay Editor are the Main Menu and Edit area. MAIN MENU The Main Menu provides access to all functions of the Hearsay Editor. This option appears in the top row of the screen. They are HELP, LOAD, SAVE, TRAIN, and QUIT. To select an option, highlight it by pressing the spacebar, direction keys, or by moving the mouse if one is in use. When the desired option is highlighted, press the [ENTER] key. The HELP Menu will appear. The options are: CANCEL - Exits from the main menu help option, & returns to the main menu. HELP SYSTEM - Is an explanation of this section. MAIN MENU - Is an explanation of the Main Menu. OVERVIEW - Is a brief description of the Hearsay Gold system. LIBRARY - Is an explanation of the Hearsay's Library system. SYSTEM USE - is a brief description of the Vocabulary & Control Windows. MOUSE USE - Is an explanation of Hearsay's use with the Microsoft Mouse. VOCABULARY - Is reserved for Hearsay's Execuware products. To select any of the above highlight the desired function & press [ENTER]. The On-Line Help System screens can also be brought up at any time by pressing the F1 key on the keyboard. The F1 key is the first function key, which is usually located at the upper left hand corner of the keyboard. To obtain information on a particular command or option, highlight the option and press the F1 key. Pressing the F1 key will not cause the loss of data during any of the modes of operation except during a train operation. In this case it would be necessary to start the training procedure over again. The messages displayed are context sensitive, and are intended to give information related to the currently highlighted command option. EDIT AREA The center of the screen is referred to as the Edit Area. The vocabulary is displayed in a layered window format. The Vocabulary created may be saved in a disk file, or into the library. If desired the Vocabulary may be erased. STEP ACTION 1. Type CD\HEARSAY [ENTER] to change to the Hearsay directory. 2. Type XXX (Your initials) [ENTER] or XXXE [ENTER] 3. Type REDT or Type MKLIB25 and then RE. 4. Select EDIT by either pressing [E] or moving the cursor over EDIT and pressing [ENTER] from the Main Menu. 5. Select VOCABULARY by either pressing [V] or moving the cursor over to VOCABULARY and pressing [ENTER]. 6. Type the vocabulary name you wish to enter when prompted at the bottom left of the screen "Select or Enter Command Name>_". NOTE: Vocabulary Names may be edited with the INS, DEL, BACKSPACE & CURSOR Keys 7. Press [ENTER]. The name will now appear in the Edit Area. 8. Hearsay will prompt you in the lower left hand portion of the screen, to say the word aloud you just typed in two or more times. NOTE 1: The Hearsay Gold should be spoken to in a similiar fashion as when speaking to a dog. A commanding authoritative tone usually results in the best performance. If Hearsay prompts you to say the word more than two times it is interpreting your second training pass differently than the first training pass. NOTE 2: You may press the [SPACEBAR] to pause the training, [ENTER] to skip or [ESC] to exit. 9. Type the Command's Keystrokes followed by the [SCROLL-LOCK] key to end the sequence. The [NUM-LOCK] key acts as a backspace if you make an error while typing. The Command Keystrokes are exactly what you wish the Hearsay Gold to do when you say that word. NOTE: Depression of the [ENTER] key within the Keystroke Sequence will not cause a new line, an [ENTER] will be executed along with any other keystrokes entered for each vocabulary word. All keystrokes including Control, Alternate, and Function keys are acceptable in a given keystroke sequence. A sequence is terminated only by pressing a [SCROLL-LOCK] key. Hearsay Gold prints the names of the keys pressed in the keystroke sequence even if they are normally unprintable. The [SCROLL-LOCK] key does not appear in the printed string because it is only a termination mark and not part of the sequence. It is important to note that the name of the vocabulary is independent of the keystroke sequence. 10. To continue entering additional vocabulary words, repeat steps 6 to 8. 11. When you are finished, to exit you must press [ENTER] then select QUIT EDIT, by either typing "Q" or moving the cursor to QUIT EDIT & pressing [ENTER]. Alternatively you can press the [ESC] key. As described later on Section 5.03 the recognition parameter DELAY COUNT is automatically calculated and the default setting of 20 must be changed to -1 for every vocabulary that you create. Do so at this point. STEP ACTION 1. Select EDIT or press E from the Main Menu. 2. Select PARAMETERS or press P from the Edit Menu. 3. Select DELAY COUNT or press D from the Edit Parameters Menu. The bottom of the screen will read: "Recognizer Delay Count (Default is 20) >20". 4. Type -1 (minus one) [ENTER] 5. Select QUIT EDIT or press Q from the Edit Parameters Menu. You now have to train, save or erase your vocabulary before exiting the Edit Area. To TRAIN see Section 2.03, To SAVE see Section 2.04, to ERASE see below. NOTE: If QUIT is selected before saving or erasing, Hearsay will prompt you for a vocabulary name and save it to the library if there is enough space. See Save Section 2.04 for choosing a legal vocabulary name. CAUTION: If Erase is selected before Saving, your vocabulary will be deleted and it will not be recoverable. TO DEFINE ANOTHER HEARSAY GOLD VOCABULARY : To define another vocabulary the Edit Area must be erased first. When the (REDT) program is run without any parameters, as in step 3 above, the edit area is automatically erased. If however, there are commands visible in the edit area, it will need to be erased before entering another vocabulary. TO ERASE THE EDIT AREA: CAUTION: Erasing a vocabulary that has not been previously saved will be deleted and unrecoverable. STEP ACTION 1. Select EDIT [ENTER] 2. Select ERASE [ENTER] or "E". If changes were made to the current vocabulary, you will be prompted to confirm the selection. If you wish to continue with the ERASE, select ERASE a second time. If you did not save the vocabulary & wish to do so, select CANCEL, & SAVE as described in Section 2.04 below. 3. Repeat steps 4 through 11 above. After defining and saving your vocabulary you must TRAIN the Vocabulary at one time for consistency. 2.03 HOW TO TRAIN A HEARSAY GOLD VOCABULARY In order for Hearsay to be able to understand spoken commands, it must have information on how the user says the words. In order to collect this necessary information, each user must "train" the system. To train a vocabulary the user is asked to say each vocabulary name two times. CAUTION: Avoid excessive background noise during training. In order to train a vocabulary it must be loaded into the Editor first. If you have just defined it in Section 2.02 you will already be in the Editor. If you saved and erased it before training you will need to load it back into the Editor by: STEP ACTION 1. Type CD\HEARSAY 2. Type XXX or XXXE 3. Type REDT VOCABULARY NAME or MKLIBXX (XX=kbytes of memory) RE VOCABULARY NAME You will now have your vocabulary ready for training. TO TRAIN: STEP ACTION 1. Type "T" for "TRAIN" or move the cursor to TRAIN and press [ENTER]. 2. Type "T" for "TRAIN WORDS" or move the cursor to TRAIN WORDS and press [ENTER]. NOTE: CANCEL - Will return you to the Main Menu. To CANCEL, either press "C" or move the cursor to CANCEL and press [ENTER]. 3. The vocabulary will be shown in the center of the edit screen. The bottom left hand portion of the Hearsay Editor will prompt you to say each word. Say the words as they appear. When the list is finished, you will be prompted to say each word again. NOTE 1: The Hearsay Gold should be spoken to in a similiar fashion as when you are speaking to a dog. A commanding authoritative tone usually results in the best performance. NOTE 2: Do not pause between a two or three word vocabulary command. NOTE 3: If the computer interprets your second training pass as considerably different than your first training pass, you will be prompted to repeat the word over again. NOTE 4: Avoid over pronunciation of words, unless this is the way you say them in normal usage. NOTE 5: To pause training press the [SPACEBAR], to skip a word press [ENTER] to exit training press [ESC]. 4. When all of the words are trained twice, the file should be saved as described in Section 2.04 below. If you do not feel that you have said the vocabulary words in a clear and consistent manner, we suggest that you select train again, and repeat steps 1 through 3 above. 2.04 HOW TO SAVE VOCABULARIES The SAVE operation allows vocabularies to be saved to the Library and permanently to a file. The vocabulary is saved to a users personalized directory (XXX) which was created in SETUP (Section 1.00). If you wish to save the file to a different directory, the drive and directory parameters should precede the vocabulary name: ie. DRIVE:\DIRECTORY\VOCABULARY NAME If the file already exists you will be able to overwrite the file with the updated vocabulary. The Vocabulary name serves two purposes in the Hearsay system. It tells the system the NAME (& optionally the location) of the file that the vocabulary data is stored in, and it serves as a LABEL to identify the vocabulary. When the vocabulary is saved to a file or Library you can choose whatever vocabulary name you like. It is recommended you use simple names such as DOS, WORDSTAR, 123, etc. If we created a vocabulary using DOS commands (The basic software that allows you to run programs, copy files, etc.), we should choose to label this this vocabulary "dos". In this case it is easy to remember the contents of the vocabulary. If the vocabulary file is located in your personal directory we could enter the vocabulary name as "dos". If the file was located on a different drive and directory, we could enter the name as DRIVE:\DIRECTORY\DOS. In this case the drive and directory would be used only to locate where to read the file from, or save the vocabulary to. The name that would appear in the Main Menu, or in the Library list would be "dos". Hearsay strips off any drive designations, extensions, or path descriptors from the vocabulary name. If a directory has been previously set up through the execution of the SETUP program and no drive or directory information is specified in the vocabulary name, Hearsay will search that directory for the file first. Each of the directories specified in the PATH will then be searched sequentially. For more information on use of the PATH command the user is referred to the DOS documentation. NOTE: The Vocabulary names are CASE SENSITIVE. Saving the vocabulary using lower case letters is not the same as uppercase. The HEARSAY GOLD distinguishes uppercase differently than lower case. For example, the name "vocabulary" is different than the name "VOCABULARY". When you want to specify a vocabulary name BE CONSISTENT with the CASE you choose. It is good practice to save your work periodically to prevent loss of data in case of power loss or accidental erasure. To SAVE select "S" or highlight SAVE and press [ENTER] from within the Hearsay Editor. The SAVE option has four options available. They are: 1. CANCEL - Will return you to the Main Menu. Either press "C" or move the cursor to CANCEL and press [ENTER]. 2. FILE - Used to save the vocabulary to the hard or floppy drive 3. LIBRARY - Used to save the vocabulary into the Library. 4. ERASE LIB Erases and removes the entire library from memory. TO SAVE VOCABULARIES TO A FILE: The user must enter a filename. The file is automatically saved to your personalized directory. If the file is not to be saved to your personalized drive and directory, a different drive and directory should precede the filename. No file extensions should be used, (ie. .EXE, .TXT). Hearsay Gold automatically appends an extension of .VOC to the filename. STEP ACTION 1. Select SAVE by either typing "S" or moving the cursor to SAVE and pressing [ENTER]. 2. Select FILE in the same manner as above. Hearsay will prompt you to enter a vocabulary name. Type the name in and press [ENTER]. NOTE: If the filename already exists, Hearsay Gold will ask you if you wish to OVERWRITE the existing Vocabulary file with the current Vocabulary file. If you are saving an existing Vocabulary, the name of the file will appear at the bottom of the screen. If you wish to save the vocabulary under the same filename, press [ENTER]. Hearsay will again prompt if you wish to OVERWRITE the old file. 3. Select OVERWRITE if you wish to save the current vocabulary file. under a different name, select CANCEL and repeat steps 1 and 2. When prompted to enter a Vocabulary name enter a new name. NOTE: HEARSAY will assign a .VOC extension to your filename and show the directory to which this file will be saved. This is your personalized directory you set up. If you wish to enter a new drive and directory do so at this point. Now that your vocabulary is saved, you may load it at any time using the REDT program described in Section 3.00. It is not necessary to retrain the vocabulary every time it is loaded. TO SAVE VOCABULARIES TO THE LIBRARY: This option places a vocabulary in the Hearsay Editor into the Library which is located in the computer's memory. If a library is not allocated or is not large enough, your file will not be saved to the Library. A message on the bottom left hand portion of the Edit Screen will say "No Space In Library, File Not Saved". Loading a large number of vocabularies into the Library will leave less room for application programs. The amount of Library space required to store a vocabulary is a function of the vocabulary size. The Hearsay system requires about 25K bytes for the 1st 64 word vocabulary and an additional 12K for each additional vocabulary. NOTE: The SAVE to the LIBRARY operation is automatically performed each time you exit the Hearsay Editor. STEP ACTION 1. Select the SAVE Option. 2. Select the LIBRARY Option. 3. Enter a vocabulary name. If the file was saved previously the name will appear at the bottom left of the screen. If you wish to save the vocabulary under the same file name press [ENTER], otherwise type in a new name and press [ENTER]. TO ERASE THE LIBRARY: As described above, when exiting the Hearsay Editor the current vocabulary in the Editor will automatically be saved to the Library if there is enough room in memory. In order to ERASE the Library one must ERASE the Edit Area first as described in Section 2.02 above. STEP ACTION 1. Select Edit 2. Select Erase. If changes were made to the vocabulary you will be prompted with a confirmation. To continue select ERASE again or CANCEL to return to the Main Menu. 3. Select SAVE 4. Select ERASE LIB 5. Select ERASE LIB again or CANCEL to cancel the operation and return to the Main Menu without erasing the Library. NOTE: If the system was installed using REDT erasing the Library also removes the space allocated for the Library for the system. If the system was installed using MKLIBXX & RE the ERASE LIB option will clear the Library however the space allocated (XX) will still be resident. 2.05 HOW TO LINK VOCABULARIES It is often convenient to segment large vocabularies into a number of smaller ones to optimize accuracy, response time & rejection characteristics of an application. To support this type of environment, Hearsay Gold provides the capability of activating a new Vocabulary from the memory Library upon the execution of a voice command. Each word in a vocabulary can activate another 64 word vocabulary in the Library. To make a voice command automatically change the active vocabulary to activate a new vocabulary follow the instructions given in the previous sections for creating commands, but DO NOT press the [SCROLL-LOCK] key to end the keystroke sequence. After a keystroke sequence for the command is entered, type [ALT-0] (This is done by holding down the ALT key on the bottom left of your keyboard then depressing the 0 key on the top row). [ALT-0] tells the Hearsay Gold system to change vocabularies. Following the [ALT-0] character, type the name of the Vocabulary to be activated. End the command sequence by pressing the [SCROLL-LOCK] key as you would to end an ordinary Hearsay Gold command. When the vocabulary word is recognized and the command is executed, the active the active vocabulary will be changed to the new vocabulary specified in the command sequence. If the requested vocabulary name is not in the Library when the command is executed, two short beeps will be heard and the active vocabulary will remain unchanged. NOTE: It is not necessary for the new vocabulary to be in the library when the command is created, only when it is executed. As an example, here is a command that will display a directory listing in DOS and then make a vocabulary called "menu", which is presumed to be in the Library, the active vocabulary: dir[ENTER][ALT-0][MENU][SCROLL-LOCK] NOTE: In this example [ALT-0] represents pressing the zero key on the top row of the keyboard while holding down the ALT key on the bottom row. The name entered should be the exact name of the vocabulary to be loaded as it appears in the Library. Whenever this command is executed, the new Vocabulary will automatically become the active Vocabulary. The LINK function is only capable of activating Vocabularies which are already in the Library resident in the computer memory. NOTE: If an attempt is made to link a Vocabulary that is not currently in the Library, the request will be ignored, and two short beeps will be heard. A vocabulary cannot be activated from disk in this manner. If an application uses several vocabularies which are linked, it is necessary to train all of these vocabularies for each new user and load all of these vocabularies into the Library. REMEMBER, vocabulary names are CASE sensitive. Trying to link a vocabulary using lowercase will not work if that vocabulary is in the Library as uppercase letters. The name "vocabulary" is different that the name "VOCABULARY". NOTE: If there is no vocabulary name specified after [ALT-0] the vocabulary that called the current vocabulary will become active. 2.06 HOW TO EXIT THE HEARSAY GOLD EDITOR AND RETURN TO DOS To return to the DOS prompt (ie. to exit the Editor program), select the QUIT option from the Hearsay Gold Editor Main Menu. If changes were made to the vocabulary you will be prompted for confirmation. To Exit select QUIT again or CANCEL to return to the Main Menu. After a short pause, the DOS prompt will return. NOTE: When Exiting the HEARSAY EDITOR the Vocabulary in the Edit Area will automatically be saved to the LIBRARY, if there is enough space & become the active Vocabulary. If this is not desired you must ERASE the Edit Area before Quitting as described above in Section 2.02.