Google announced speech translation across 63 languages in Meet on October 22, 2026, including automatic language detection and synchronized translated captions. Rapid Release rollout starts October 22; Scheduled Release starts November 2, with eligible plans and updated mobile apps required. This update concerns translation, not verified speaker identity. If you later build minutes from authorized audio or a transcript, keep “what was said” separate from “who said it.” Do not treat translated live captions as the original-language transcript.
For the example below, use a transcript from a tool that actually provides speaker labels; check that capability separately from Meet’s translation feature.
A transcript that separates speakers does not automatically identify those people. A label such as Speaker A means a voice segment was assigned to a group; attaching a real name requires evidence.
Build the name mapping separately
For an authorized recording, retain segment timestamps and a table of label, proposed name and supporting passage. A spoken introduction can help, but check the voice across later turns. Overlapping speech, brief interruptions and similar voices deserve manual listening.
For example, Speaker A says “I’m Mira” at the start. Verify a later clear segment before renaming every A segment. If the mapping is uncertain, leave the neutral label rather than guessing from job roles.
Check model and privacy limits
Some transcription workflows support speaker reference clips; check current support and permission to use the recordings. The compatibility examples for gpt-4o-transcribe-diarize have a February 26, 2027 shutdown date. They are not a permanent recommendation for a new integration; check the current model’s actual speaker-label support separately. Before sharing, verify names against the audio, not the fluency of the transcript.