Google announced today that the Gemini Audio suite now includes Gemini 3.5 Transcribe, an AI‑driven speech‑to‑text model that automatically strips filler words such as “um” and “uh,” detects specialized jargon, and supports more than 85 languages. The feature rolls out in English for macOS Gemini app users, appears in the Rambler dictation feature on Android in selected markets, and is available to developers through a public preview of the Gemini API.
Key Takeaways
- Gemini 3.5 Transcribe removes filler words and formats text automatically.
- The model recognizes over 85 languages and can be customized with user‑defined vocabularies.
- It attributes speech to up to three speakers and provides word‑level timestamps.
- Available today on macOS, Android (limited regions), and via the Gemini API; Chrome support is forthcoming.
| Metric | Value | Why it matters |
|---|---|---|
| Product | Gemini 3.5 Transcribe | First Gemini Audio model to edit filler words automatically. |
| Languages supported | > 85 | Broadens global usability for multilingual teams. |
| Speaker attribution | Up to 3 speakers | Enables clearer meeting minutes and multi‑speaker podcasts. |
| Release date | Rolling out today (August 26 2024) | Marks the first public availability of the new model. |
What Gemini 3.5 Transcribe Does
The new model builds on Google’s earlier Chirp 3 transcription engine, which it describes as a “major advancement” in multilingual performance and wording error rates. By automatically removing filler utterances and applying natural‑language formatting, the service lets users edit documents simply by speaking. In addition, developers can upload a custom vocabulary so the model respects domain‑specific spelling and terminology without manual post‑processing.
Technical Improvements Over Chirp 3
Google claims that the shift from Chirp 3 to Gemini 3.5 Transcribe yields lower word‑error rates across the expanded language set. The model also introduces word‑level timestamps, a feature that facilitates precise alignment of transcript text with audio playback—a critical capability for subtitle generation and searchable archives.
“Gemini 3.5 Transcribe represents a major advancement from our previous transcription model, Chirp 3,” Google said.
Customization and Speaker Identification
Beyond filler removal, the service can differentiate up to three speakers in pre‑recorded audio, tagging each segment with the appropriate speaker label. When paired with a custom vocabulary, the model adapts to industry‑specific jargon—preventing unwanted auto‑corrections that often plague generic speech‑to‑text solutions.
Rollout Strategy and Platform Availability
Google is deploying Gemini 3.5 Transcribe in stages. The first wave reaches all macOS users of the Gemini app in English, while Android users gain access through the Rambler dictation feature in a limited set of countries and languages. Developers can experiment with the model via a public preview in the Gemini API, accessible through AI Studio and the Antigravity console. Chrome integration is slated for a future update.
Related Gemini Audio Updates and the Pro Model Delay
In the same announcement, Google referenced upcoming Gemini 3.5 Live and Gemini 3.5 Live Experimental models that would enhance real‑time speech recognition, interruption handling, and visual processing. However, after the story was published, Google clarified that those models are not launching today and provided no new release timetable. The company also remains pending on the Gemini 3.5 Pro model, which was originally promised for June.
Why it matters
For enterprises, content creators, and developers, the ability to generate clean, filler‑free transcripts in a single pass reduces editing overhead and accelerates workflows. Multilingual support across more than 85 languages positions Gemini 3.5 Transcribe as a competitive alternative to existing cloud transcription services, especially for teams that need custom terminology handling. The speaker‑attribution and timestamp features further extend its utility to podcast production, meeting summarization, and accessibility compliance.
Frequently asked questions
Which platforms currently support Gemini 3.5 Transcribe?
The model is live for macOS users of the Gemini app (English only) and for Android users through the Rambler dictation feature in selected regions. Developers can access it via the Gemini API public preview.
Can the model handle domain‑specific terminology?
Yes. Users can upload a custom vocabulary, allowing the transcription engine to recognize and retain specialized jargon without manual correction.
How many speakers can the system differentiate?
Gemini 3.5 Transcribe can attribute speech to up to three distinct speakers in pre‑recorded audio files.
When will Chrome support be available?
Google has announced that Chrome integration is forthcoming but has not provided a specific launch date.
Is Gemini 3.5 Pro still expected?
The Gemini 3.5 Pro model, originally slated for a June release, remains unreleased, and Google has not issued a new timeline.
Bottom Line
Google’s Gemini 3.5 Transcribe delivers a streamlined transcription experience by automatically removing filler words, supporting a vast multilingual roster, and offering custom vocabularies and speaker attribution. While the rollout is limited to macOS, select Android markets, and developers via API, the feature signals a clear push toward more polished, production‑ready speech‑to‑text solutions from Google.
We’ll monitor upcoming updates to Gemini 3.5 Live, the pending Pro model, and the Chrome extension to see how Google’s audio ecosystem evolves.
Reporting based on: original source.
Comments
Post a Comment