Google Unveils Gemini 3.5 Transcribe for Advanced Audio Processing

Google has launched Gemini 3.5 Transcribe, a specialised speech-to-text model designed to convert spoken audio into structured, formatted text while improving the accuracy and readability of transcripts.

The new model is designed to automatically filter out filler words, handle mid-sentence self-corrections and recognise specialised terminology and jargon. Google is positioning Gemini 3.5 Transcribe as a solution for both real-time voice applications and recorded-audio transcription.

Two APIs for Different Use Cases

Gemini 3.5 Transcribe is available through two developer interfaces: the Live API and the Interactions API.

The Live API is designed for continuous, two-way audio streaming, delivering sub-second latency for applications such as voice agents and real-time conversational experiences.

The Interactions API is targeted at recorded audio and includes features such as speaker attribution and word-level timestamps, giving developers greater control over how transcripts are processed and presented.

Supports More Than 85 Languages

The model supports more than 85 languages, including regional accents and dialects. It can also detect languages in real time.

For recorded audio, Gemini 3.5 Transcribe can identify up to three distinct speakers, while support for larger groups remains experimental.

Performance benchmarks from Artificial Analysis recorded a Word Error Rate (WER) of 4% for streaming applications and 2.6% for non-streaming use cases. On the FLEURS benchmark, Google reported WER figures of 5.50% and 5.04%, respectively.

Google also said the model reduces final transcript delivery times by 70% compared with its previous Chirp 3 model.

Integration Across Google Products

Google is incorporating Gemini 3.5 Transcribe into several products and services across its ecosystem.

On Android, the technology supports Gboard’s “Rambler” feature, which enables users to edit text through voice commands. Google Antigravity also uses screen context to improve transcription accuracy when dealing with elements such as file names and documents.

The technology additionally powers voice-driven workflows in the Gemini macOS desktop app and voice-assisted application development in Google AI Studio. Google is also preparing web-field dictation capabilities for Google Chrome.

Availability

Gemini 3.5 Transcribe is currently available in public preview through Google AI Studio and Google Antigravity, as well as through the enterprise-focused Gemini Enterprise Agent Platform.

The Gemini macOS desktop integration is currently available in English, while Gboard’s Rambler feature has begun rolling out on Android in select regions.

With its combination of real-time streaming, multilingual support, speaker identification and context-aware transcription, Gemini 3.5 Transcribe expands Google’s speech AI capabilities across developer tools, consumer applications and enterprise workflows.