Voice command library for mobile app

Proof-of-concept app description Use the CMUSpinx toolkit ([login to view URL]) , available for both Linux and Windows, which combines elements of speech recognition, to build a locally-stored command library of commands. The app will be use to incorporate into an in-development app used by taxi drivers. We want to create a proof-of-concept command-recognition function that learns 4 commands: Accept, Decline, Yes, and No. There are two phases involved in the use of the test app: a command learning phase, and a command use phase. The learning phase can take place while using the app, so no separate learning function is required to build the command library. The function works only when the taxi is in the foreground. We will provide you with the taxi app, for you to use as a test bed. When first launched the command library is empty. The Command Recognition function (CRF) detects this condition and it automatically switches to learning mode, but will only listen at specific time windows, when an event occurs on the Driver app. Assume a request for a pickup is pushed to the app from the taxi gateway. This results in an audible signal:”Ping, Ping”, on the device, a text message with the details, and a floating view, similar to that displayed when receiving a phone call while another app is in the foreground, that displays a message and two buttons. The text message duplicates the contents that appear in the view, in case the phone was turned off when received. The message is “You have a request, 1.5 miles away, for 2 passengers, estimated fare is $74.00, and trip distance is 15 miles”. And the two buttons labeled Accept and Decline. So far the app behaves no differently than the un-assisted version. The VCR, running in the background, detects the pickup event and goes into listening mode. Say the driver says Accept. Using voice recognition the app determines the Hamming difference between the spoken command and one of two possible words: Accept and Decline. Even with a strong accent Accept will have a lower score, so it is stored in the command library. Now we have one word in our library. To make sure, the CRF asks “Please confirm you accept, Yes or No?” (A similar question is always asked when in learning mode.) This view only displays if CRF is active (it can be activated in Settings). Again, we repeat the same process and store the answer. If we recognize NO instead of yes, one of two possible explanations are possible: 1) the driver changed his mind; 2) the background noise is so bad that the word Accept was mistaken for Decline. In both cases we return to the original view and redisplay the request message. We also delete the stored answer in our command library. Once a command has been stored in the local command library the app does not ask for confirmation. It would be nice to save the command library to the server whenever the app is shut down, just in case the local memory becomes corrupted and it becomes necessary to restore the library, but we can worry about this later.