Generating and/or utilizing voice authentication biasing parameters for assistant devices

A method implemented by one or more processors, the method comprising:

• receiving audio data that captures a spoken utterance, the audio data being detected via one or more microphones of an assistant device;
• generating a first user voice authentication measure, for a first registered user of the assistant device, wherein generating the first user voice authentication measure comprises:
• processing at least part of the audio data, using a speaker authentication model, to generate utterance features that correspond to the at least part of the audio data;
• and
• generating the first user voice authentication measure based on comparing the utterance features to first speaker features stored in association with the first registered user;
• determining a first user biasing parameter that is specific to the assistant device and that is specific to the first registered user, wherein determining the first user biasing parameter that is specific to the assistant device and that is specific to the first registered user comprises:
• processing one or more current contextual features using a machine learning model to generate output that reflects the first user biasing parameter and one or more additional biasing parameters for one or more additional registered users;
• wherein the machine learning model is trained based on training instances that are each based on a corresponding past interaction with the assistant device, the training instances each including training instance input that reflects corresponding contextual features at a time of one of the corresponding past interactions and ground truth output that reflects which, of the first registered user and the one or more additional registered users, was verified for the corresponding past interaction;
• generating a modified first user voice authentication measure as a function of the first user voice authentication measure and the first user biasing parameter;
• and
• determining, based on the modified first user voice authentication measure, whether to authenticate the spoken utterance as being provided by the first registered user.

A method implemented by one or more processors, the method comprising:

• receiving audio data that captures a spoken utterance, the audio data being detected via one or more microphones of an assistant device;
• generating a first user voice authentication measure, for a first registered user of the assistant device, wherein generating the first user voice authentication measure comprises:
• processing at least part of the audio data, using a speaker authentication model, to generate utterance features that correspond to the at least part of the audio data;
• and
• generating the first user voice authentication measure based on comparing the utterance features to first speaker features stored in association with the first registered user;
• determining a first user biasing parameter that is specific to the assistant device and that is specific to the first registered user, wherein the first user biasing parameter that is specific to the assistant device and that is specific to the first registered user is based on a calendar entry that indicates the first registered user is anticipated to currently be in a room to which the assistant device is assigned;
• generating a modified first user voice authentication measure as a function of the first user voice authentication measure and the first user biasing parameter;
• and
• determining, based on the modified first user voice authentication measure, whether to authenticate the spoken utterance as being provided by the first registered user.