Google's new AI removes 'ums' and 'ahs' from your speech
What's the story
Google has unveiled its latest artificial intelligence (AI) model, Gemini 3.5 Transcribe, which is touted to be the company's most accurate speech-to-text engine yet. The innovative system is capable of handling background noise, technical jargon, and filler words with unprecedented efficiency. It also adapts to individual vocabularies and speaking styles for more seamless edits.
Enhanced features
Can handle up to 3 speakers at a time
Gemini 3.5 Transcribe is already being used in the Gboard Rambler feature, and it will soon enable voice-to-text functionality directly in Chrome.
The model supports over 85 languages, can recognize different accents, and handle up to three speakers at a time with timestamps.
It is also significantly faster than its predecessor, processing audio data 70% quicker while maintaining a low word error rate of just 4% for streaming and an impressive 2.6% for non-streaming tasks.
Expansion strategy
Gemini 3.5 Transcribe coming to more devices later this year
Currently, you can use Gemini 3.5 Transcribe on Mac or via Google Antigravity, with more platforms like Chrome set to support it soon.
The model is already live in some places, including Rambler in Gboard for Pixel 11 phones.
Google plans to expand this feature to "more Gemini Intelligence devices" later this year, showing its commitment to making advanced AI transcription technology widely accessible across its ecosystem of products and services.