Google launched Gemini 3.5 Transcribe for speech transcription. Belarusian is on the list
Google has introduced Gemini 3.5 Transcribe — a new speech recognition model. It supports over 85 languages, including Belarusian, and can clean text from filler words and self-corrections, writes Dev.by.

Unlike regular transcription, the model doesn't just literally record audio. For example, if a user says: "Let's meet on Tuesday — no, on Wednesday", the corrected version will remain in the final text. Gemini 3.5 Transcribe also automatically formats text and can take into account a user dictionary — for example, professional terms, names, or unusual spellings.
Two modes are available for developers. Gemini 3.5 Transcribe Live processes spoken language with a delay of less than a second and is designed for voice assistants, subtitles, and other real-time applications. The standard Gemini 3.5 Transcribe works with ready-made audio recordings — meetings, calls, and other files — and can add timestamps and identify speakers.
In recorded audio, the model can distinguish up to three interlocutors. Support for a larger number of speakers is currently in experimental mode.
According to independent service Artificial Analysis, the average rate of incorrectly recognized words in Gemini 3.5 Transcribe is 4% for streaming processing and 2.6% for working with ready-made recordings. Google also claims that the time to obtain a final transcription has been reduced by 70% compared to the previous Chirp 3 model.
Gemini 3.5 Transcribe is already used in several Google products. In the Gemini app for macOS, it allows dictating text, removing pauses and filler words, and executing voice commands taking into account screen content. In Gboard, the model underlies the Rambler feature, which converts free speech into formatted text and allows voice correction of what is written or changing the style.
In Google Antigravity, the model can consider screen content and chat history to more accurately recognize, for example, file names and terms from open documents. Developers can already try Gemini 3.5 Transcribe in public preview via the Gemini API in Google AI Studio and Google Antigravity. For enterprise customers, the model is available through the Gemini Enterprise Agent Platform.
Now reading
"For the whole world to speak Russian, study the Kalashnikov assault rifle." In Russia, a war participant gave an order to schoolchildren on September 1
Comments