ASR tests isolating one speaker in overlapping speech

Published on September 15, 2026, Xiaomi-CocktailASR-1 is presented as a demonstration for transcribing only one speaker when speech overlaps. Its workflow uses an audio sample of that person to try to isolate their voice; engineers can test this speaker-conditioned speech recognition.

The description provides no metrics, evaluation results, or limitations, so it does not establish how well the approach works. Consult the demonstration and verify it independently using authorized recordings and reference transcripts. If using AI to study or test the technique, limit audio to what is necessary and protect identifiable recordings according to your organization’s rules.