Should You Buy Speech Recognition from Your Simulation Vendor?
CAE A350 Level D Simulator

Should You Buy Speech Recognition 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 Automatic Speech Recognition (ASR) is a valuable addition to human-in-the-loop simulators. ASR brings many benefits not least being that it can significantly reduce the cost of the human role-players found in many training applications. The upfront initial investment for a capable and usable ASR is quickly returned. This is a feature where shortcuts and cost-cutting will most definitely leave you with an expensive and unusable speech capability.

When correctly designing and implementing an ASR into your training system for the first time, the cost is not insignificant. 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. Acoustic model tuning will be required to achieve the best performance and the ASR must be provided with information that does not exist in a traditional simulator, 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.

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. Ongoing adaptations to the acoustic models will be necessary.

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?

If the answer to either of the above scenarios is yes (or maybe) then it is important to establish from your vendor how either task can be achieved. The following questions for your simulation vendor will give you some indication of the ease or difficulty in doing either:

  1. Is the ASR offered in your solution standards-based?
  2. Does the grammar support W3C SRGS and SISR? Can the grammar be reused with an alternate ASR vendors application?
  3. Is the acoustic model proprietary? What capabilities exist to adapt the model for use with another ASR?
  4. Do you have an open API/ICD for integration with the ASR component if we choose a new simulator?
  5. 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. If this feature is tightly coupled, the ASR will not function in the same manner on any new simulator.
  6. 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?
  7. Is domain-specific data (phonetic spoken names) available in a standalone reusable file?

An ASR for your training system will serve you well for a very long time. It is sensible that with the investment of time, money, and knowledge into the development of the capability, you plan the possible transfer of the ASR to a new simulator. Additionally, if you determine a new speech technology is required, how many of the simulator provided speech components might be usable for the replacement project. Why would you leave yourself in the position of once again having to invest heavily to achieve similar capabilities?

When planning for a new ASR enabled simulator or the addition of ASR to your existing simulator, take the time to think about an architecture that will minimize the cost of doing both. An ASR is certainly a component that can be bought from an independent vendor. As a minimum, you should seriously consider a design where the ASR is a standalone component. In your requirements specification, it would be advisable to clearly define the requirements to treat ASR as portable or replaceable. (See this free download for a helpful guide Writing ASR Requirements. Portability provides flexibility. If you do choose to buy from your simulator provider, you provide an exit path to a new simulator should it be desired. You have the option to take your ASR with you.

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.

To view or add a comment, sign in

More articles by Gary Pearson

Insights from the community

Others also viewed

Explore topics