A method comprises capturing, by a microphone of each of a plurality of conference computing devices, multiple distinct spoken utterances. The utterances are provided by multiple participants of a meeting facilitated by the conference computing devices. Speech-to-text processing is performed on the utterances to generate multiple speech recognition outputs, upon which semantic processing is performed. Based on the semantic processing, a plurality of responsive outputs that correspond to the multiple distinct spoken utterances are generated. The method further comprises identifying respective output modalities used by the plurality of conference computing devices, and selecting respective subsets of the plurality of responsive outputs to be output at respective conference computing devices. The method further comprises outputting a respective subset of the plurality of responsive outputs at each conference computing device.