Are you buying a simulator with Automatic Speech Recognition (ASR)? Are you adding speech recognition to your existing simulation capabilities? The ASR buying decision is already complex. (see this Free Download). When choosing a speech recognition vendor, purchasers often overlook this key question. Should you buy ASR from your simulation vendor?
If you are looking to save time and avoid reading to the end, the answer is sure why not……But.
A well-designed and well-implemented ASR is clearly a valuable addition to many human-in-the-loop simulators. ASR brings many benefits not least being that it can significantly reduce the cost of the human role-players. You will see a return on investment by taking the time with design and planning. ASR is a feature where shortcuts and cost-cutting will leave you with an expensive and unusable speech capability.
ASR Integration Can Be Complex and Costly
When correctly designing and implementing an ASR the cost can be significant. Some applications require support for tens of thousands of different phrases. Simulated entity responses require multiple TTS licenses. Simulators will typically require modifications to support the instructions returned by the ASR. Best performance requires ongoing acoustic model tuning. Information that does not exist in a traditional simulator, must be defined e.g., in aviation-related applications, the phonetic names for aircraft types, navigation beacons, and airline designators.
If you intend to add ASR to your training capabilities there is no option but to plan in detail and budget accordingly. When choosing a speech recognition vendor, the decision should in part be influenced by your future plans.
Making the Speech Recognition Buying Decision, Why Does it Matter?
ASR as stated is a worthwhile investment of time, knowledge, and money. After the initial purchase, ongoing improvements are necessary. They include adding to the supported phraseology & spoken name entries in the dictionary and ongoing adaptations to the acoustic models.
Is it possible that you may consider replacing your simulation technology and the associated vendor?
Alternatively, is it possible that new speech technology will lead you to consider keeping your simulation technology and replacing the ASR?
Making a Decision
If the answer to either of the above scenarios is yes (or maybe) then questions must be asked of your potential vendors. It is important to establish how either replacing the speech and keeping the simulator or replacing the simulator and keeping the speech can be achieved. The following questions will guide you in finding the answer to the question, should you buy ASR from your simulation vendor?
- Is the ASR offered in your solution standards-based?
- Does the grammar support W3C SRGS and SISR? Can the grammar be reused with an alternate ASR vendors application?
- Is the acoustic model proprietary? What capabilities exist to adapt the model for use with another ASR?
- Do you have an open API/ICD for integration with the ASR component if we choose a new simulator?
- Are any parts of the ASR solution reliant on tight integration with your simulator, e.g., a well functioning ASR will have some form of validation of the ASR result before processing by the sim-engine. On a new simulator, the ASR will not function in the same manner if this feature is tightly coupled,.
- What components of the ASR integration exist outside of the simulator (e.g., command post-processing, pilot/controller A.I.), and are these components documented and reusable?
- Is domain-specific data (phonetic spoken names) available in a standalone reusable file?
An ASR will serve you well for a very long time. Having the ability to transfer the ASR to any new simulator is important. How many of the simulation speech support components might be reusable? Why would you leave yourself in the position of having to invest again to achieve similar capabilities?
So, Should You?
Take the time to think about an architecture that will minimize the cost of exchanging either the ASR or the simulator. You can certainly buy an ASR from an independent vendor. In your specification, you must clearly define the requirements to treat ASR as portable or replaceable.
So, should you buy your ASR from your simulator vendor? Sure why not……But make sure that they can provide satisfactory responses to your portability questions.
