Adrian Als - Sheffield Hallam University
Error Correcting codes for voice generated cryptographic keys
We are exploring a myriad of error correcting codes for protecting voice-generated cryptographic keys. A primary concern of our work is in developing strategies for correcting errors that arise in a speakers acoustic environment while she utters her password. These errors arise for a number of reasons, including but not limited to, noise, poor microphone quality, distance from the microphone, and changes in the speakers vocalization of her password. To accommodate for these errors, when reconstructing a speaker's derived key, it is necessary to try alternate feature descriptors that we deemed ``close'' to the speakers extracted feature descriptor. These alternative feature descriptors are derived by various error correcting strategies, and the number of alternatives that can be accommodated is dictated primarily by the computation time we are allowed to spend before returning a negative answer to the user. As such, the error correcting strategy that we apply needs to be as efficient as possible.