MSDN Magazine - April 2008 - (Page 61)

Figure 3 Prompts with Extraction Annotated The process of creating and importing these WAV files can be tedious if your application has numerous prompts. Let’s consider the following prompts: • “I’ll transfer you to a representative.” • “I’ll transfer you to a loan officer.” • “A representative will be with your shortly.” • “A loan officer will be with you shortly.” Instead of creating four different prerecorded prompts you could use an extraction technique. Extraction allows you to combine all or part of a phrase from one WAV file with all or part of a phrase from another. Extraction doesn’t happen automatically; you must specify in the transcription field which part of the transcription can use extraction. So instead of having four different WAV files in the prompt database you can simply have two WAV files with extraction brackets, as shown in Figure 3. When the application encounters the phrase “I’ll transfer you to a,” it will automatically combine that specific audio portion with another portion of audio. It will even mix prerecorded prompts with TTS if part of the encountered prompt isn’t in the prompt database. So if you later add another prompt that says, “I’ll transfer you to a mortgage specialist,” the phrase “I’ll transfer you to a” will use the WAV file, while the “mortgage specialist” phrase can be rendered using the TTS engine. Keyword and Conversational Grammars Now that your application can talk, you’ll want it to be able to accept responses from the user either through speech recognition or DTMF. Speech Server 2007 does not automatically recognize everything the user says, so you must specify acceptable user responses. These are defined in your application’s grammar. In a speech recognition app, there are two types of grammars: keyword and conversational. Keyword grammars are based on the use of specific words in the response. Keyword grammars are good for asking the user a very direct question, such as, “Do you want Account Information, Loan Inquiry, or to speak with a representative?” Figure 4 Conversational Grammar Builder If the user says anything other than the list of acceptable answers, even if the user’s response is very similar to what the application is looking for, the application would probably respond with something like, “I didn’t understand what you said. Please say …” Conversational grammars, on the other hand, try to address this problem by taking some of the responsibility off the user and putting it back on the application to figure out what the user wants. Instead of asking a direct question, you ask the user an open-ended question: “How can I help you?” This is a more natural way of asking a question, and you will probably get a more natural and complex answer back from the user. The Conversional Grammar Builder allows you to build conversational grammar easily, taking much of the effort out of working with a statistical language model yourself. The Conversational Grammar Builder, shown in Figure 4, is divided in two sections: Keywords and Answers. In the Answers section, you have three options: Concept Answer, Keyword Answer, and Command. Concept Answers differ from the Keyword Answers in that Keyword Answers require the caller to say the exact phrase to trigger recognition while Concept Answers allow the user to give a response similar to one of the Answer phrases you have predefined. You’ll want to add one Concept Answer for the main menu prompt for this example (see Figure 4). The Concept Answer typically represents one question you are asking the user to give a response to, in this case the main menu question, “How can I help you?” Next, you’ll need to add one Concept to the Concept Answer for each of the possible choices the application will allow for this prompt. For the Main Menu QuestionAnswer activity created earlier, I would have one Concept Answer named MainMenu and three concepts named AccountInformation, LoanInquiry, and Representative. Now consider this phrase: “I’d like to pay my loan.” If this phrase Voice Response Workflows april2008 61

Table of Contents Feed for the Digital Edition of MSDN Magazine - April 2008

MSDN Magazine - April 2008 Contents Toolbox CLR Inside Out Basic Instincts Cutting Edge Foundations Test Run Service Station Windows with C++ Going Places { End Bracket }

MSDN Magazine - April 2008

For optimal viewing of this digital publication, please enable JavaScript and then refresh the page. If you would like to try to load the digital publication without using Flash Player detection, please click here.