Activation word at the beginning,
at the end
More flexibility for your users! By buffering the incoming audio (from the microphone) on the local system, the virtual assistant has access to the entire utterance or command including the beginning spoken before the activation word.
In the example, “Give me the weather forecast” has been spoken before the activation word “Abbie,” and because of the local buffer (e.g., a ring buffer), the whole utterance can be streamed to the voice recognition system in the cloud for processing including the first part, “Give me the weather forecast,” the activation word, “Abbie,” and the second part, “for the weekend.” The beginning of the utterance or command can be detected by means of a speech pause or silence.
This gives your users more flexibility in articulating a statement but still preserves their privacy because the audio is only buffered locally.