Query Understanding and Voice Interfaces
Part of the Query Understanding series by Daniel Tunkelang.
Overview
Voice search differs from typed search in ways that require the entire query understanding pipeline to adapt. Voice queries are longer and more conversational in style, reflecting the natural tendency to speak in complete sentences rather than keyword fragments. The system must also handle a distinct error type: mishearings and substitutions introduced by speech recognition, which don’t follow the same patterns as typed spelling mistakes and require different correction approaches. Critically, voice interfaces can only return a single spoken answer rather than a ranked list of results, which raises the stakes for getting the most likely interpretation right and for gracefully communicating uncertainty. The growing use of voice for ambient and hands-free contexts has introduced query types with distinctive characteristics that desktop search never encountered.
Note: Article content behind Medium paywall — accessible at source URL with Medium account.