Google has introduced Gemini 3.5 Transcribe, an AI audio model that the company says improves speech recognition, transcription accuracy and multilingual speech-to-text.
Google said the model can automatically detect more than 85 languages and convert unstructured speech into formatted text. The company also said users can edit text with voice commands, while the system removes filler words from transcripts.
In a post on X, Google said Gemini 3.5 Transcribe “seamlessly handles self-corrections,” removes filler words and understands a user’s natural intent and speaking style. The company said the model accurately captures audio as users speak.
Google said the model can learn custom vocabulary and unique spellings. It also said Gemini 3.5 Transcribe is designed to capture alphanumeric strings such as order numbers and postal codes.
The company said the system can attribute speech to as many as three speakers in pre-recorded audio and provide word-level timestamps. Google said that feature can be used for audio such as podcasts.
Gemini 3.5 Transcribe already powers the Rambler feature on Android devices including Pixel 11-series phones, and the Gemini app on macOS. Google said the macOS app can combine the transcription model with other Gemini models to perform agentic tasks.
Google said users will soon be able to use speech-to-text in any field on a web page in Chrome. The company said that will let users dictate replies and posts, and use voice prompts with Gemini.
Google said Gemini 3.5 Transcribe is also available in Antigravity, its agentic development platform, and will be added to Search Live, Gemini Live, Docs, Keep and Gmail. Developers will also be able to access the model through APIs.
Featured image credit
