Google Launches Speech-to-Retrieval for Direct Spoken Query Search

Google has introduced Speech-to-Retrieval (S2R), a new approach that maps spoken queries directly to embeddings for information retrieval, bypassing traditional speech-to-text conversion. This method improves accuracy by focusing on retrieval intent rather than transcript fidelity, using a dual-encoder architecture trained on paired audio and document data. S2R outperforms previous cascade models and is now in production, supporting multiple languages and open-sourced datasets for benchmarking.