"Voice command" does not equal "Speech recognition". That is, Build Logger supports a very limited set of sounds ("words") that it recognizes as command tokens. When the right set of tokens has been provided, an action is performed.

It is important to know when to interpret the stream of tokens. People get.. ooo a bird... interrupted all the time. Some even exclaim loudly when a solder blob misbehaves mid-command. As such, instructions need prefix and suffix markers. Tokens/words can be ignored until a prefix is found, such as "Command". Then the words accumulate until a suffix is found. Naturally, the accumulator is cleared if too much time passes or too many words are spoken.

A resulting command structure could be:

Command session - begin a build session with associated reminders to collect data

Command stop session - end the build session. Can also happen due to inactivity.

Command record - begin audio recording for use during audio tours and later note-taking.

Command timer - start a count-up timer without a limit.

Command stop - stop the current recording or timer. If a timer is stopped, its timer's final minute duration is spoke.

Command status - speak the timer's current minute duration (and any other short status information)

Command [numbers] timer - start a count-down timer for the given minutes duration.