Google Research has developed a new Speech-to-Retrieval (S2R) engine that bypasses the need for text transcription in voice searches. This approach directly interprets spoken queries to understand user intent, aiming to improve accuracy and speed by avoiding errors that can occur during speech-to-text conversion. To support this advancement, Google is also open-sourcing the Simple Voice Questions (SVQ) dataset, which includes audio questions in multiple languages. AI
Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →
RANK_REASON Google Research blog post detailing a new approach to voice search and releasing a supporting dataset.
