Google launches Gemini 3.5 Transcribe with smarter speech-to-text and 85+ language support

Google has introduced Gemini 3.5 Transcribe, which it describes as its most precise speech-to-text model yet. The model can clean up disfluencies, remove filler words, handle self-corrections and automatically format transcribed text. It can also recognise custom vocabulary, specialised jargon and unique spellings.

Must Read: Claude Cowork gets shared memory with chat: What it means for you

Turning natural speech into clean, usable text can be difficult when conversations involve accents, background noise, specialised terms or people correcting themselves mid-sentence. Google is now addressing those challenges with a new model focused on making speech transcription more accurate and useful across a range of situations. 

Gemini 3.5 Transcribe focuses on accuracy

The technology is aimed at developers building voice agents and other audio-based applications, while Google is also bringing related capabilities to several of its own products. For users, the changes could make speaking to AI more practical by reducing the need to manually correct transcripts and helping spoken instructions translate into more structured text.

Gemini 3.5 Transcribe supports more than 85 languages, including accents and dialects. For recorded audio, it can identify up to three speakers and provide word-level timestamps. Support for more than three speakers is currently experimental.

Must Read: OpenAI reveals how Jalapeño Chip performs, wider deployment coming in 2027

Google cites Artificial Analysis measurements showing a 4% average Word Error Rate (WER) for streaming and 2.6% for non-streaming use cases. Google also says the time to final transcription is 70% faster than Chirp 3. On the FLEURS benchmark, the model recorded 5.50% WER in streaming and 5.04% in non-streaming tests.

APIs and Google product integration

Developers can access the model through two APIs. The Live API supports continuous bidirectional streaming with sub-second latency, while the Interactions API is designed for recorded meetings, call logs and other audio, including speaker attribution and timestamps.

Google is also integrating the technology across its ecosystem. On Android, Gboard's Rambler can convert speech into formatted text and lets users make corrections or change writing style using voice. Google Antigravity can use screen context and chat history, with permission, to improve transcription accuracy.

In Gemini's macOS app, voice commands can analyse local files, repurpose text, generate images and complete other workflows. Google AI Studio supports voice-based app building, while Chrome will soon support dictation into web fields.

Gemini 3.5 Transcribe is available in public preview through Google AI Studio and Google Antigravity. Enterprise access is available through Gemini Enterprise Agent Platform, while the Gemini macOS app supports it in English, and Rambler is available on Android in select countries and languages.