Skip to content
News Google Blog Aug 2026

Google: Gemini 3.5 Transcribe brings intelligent speech-to-text

Google announced Gemini 3.5 Transcribe on August 26, 2026, in a blog post by Diego Melendo Casado and Luke Leonhard from the Gemini Audio team. The model converts audio into polished, formatted text with capabilities that go beyond simple transcription.

The headline accuracy figure is a 2.6% word error rate on pre-recorded audio across more than 85 languages, with automatic language detection. In streaming mode the rate is 4.0%, which is still a significant improvement over the previous Chirp 3 model and delivers text about 70% faster. The model handles background noise, overlapping speech, and technical terminology without special configuration.

Beyond raw transcription, several features address the specific needs of writers and journalists. Filler word removal — eliminating “um,” “uh,” and mid-sentence self-corrections — produces cleaner drafts from recorded interviews without manual cleanup. Multi-speaker diarization identifies up to three speakers with timestamps, which is useful for roundtable discussions, interview recordings, and panel audio. Custom vocabulary support means product names, technical terms, and unusual proper nouns get spelled correctly without post-processing.

Two deployment options cover different use cases. A pre-recorded API handles audio files up to one hour with full diarization and word-level timestamps. A streaming API enables real-time transcription with sub-second latency, suitable for live captioning and in-session note-taking. Both versions support function calling to delegate complex tasks to other Gemini models, which opens paths for more automated downstream processing of transcribed content.

The model is available now in Google AI Studio and the Gemini Enterprise Agent Platform. Consumer rollout is happening through Gboard, the Gemini app, Chrome, and Google Antigravity.

For journalists and writers who rely on recorded interviews, this reduces one of the more time-consuming steps in a typical research process. The combination of high accuracy, speaker identification, and filler removal covers most of what professional transcription services provide, at a fraction of the cost.