Add options for handling unclear/invalid responses

[becky-gilbert]

We should figure out how to handle the following cases:

• Transcription fails to detect any speech (I'm not sure if there are multiple ways for this to fail, e.g. no speech sounds detected vs inaudible speech - need to check the possible outputs of the model)
• Transcription succeeds but does not contain a valid response. This should probably be an optional 'validation function' parameter, that checks the transcription result and returns true for valid and false for invalid. For instance, a researcher may want to ensure that the transcription contains at least one word from a set of valid responses (e.g. "yes" or "no", color names, animal names). Then we may also need a parameter for handling invalid response cases, e.g. message to user, repeat trial.