Google Previews Gemini 3.5 Transcribe for Intelligent Real-Time Speech-to-Text

Google has officially introduced Gemini 3.5 Transcribe, a new speech-to-text model specifically designed for precise and intelligent real-time transcription. [4] The system converts raw audio directly into auto-formatted text by seamlessly handling speaker self-corrections and removing filler words such as "ums" and "ahs." [1][4] Gemini 3.5 Transcribe already powers the new Rambler capability on Android devices as well as the Gemini application on macOS. [4] Furthermore, the speech-to-text model is coming to the Chrome browser and is currently available in public preview for both developers and enterprises. [3]

Developers can access Gemini 3.5 Transcribe through the Gemini API in Google AI Studio and the Gemini Enterprise Agent Platform. [4] To support different developer workflows, the model is available across two separate application programming interfaces. [4] A real-time streaming API, specifically named gemini-3.5-transcribe-live, provides continuous, bidirectional streaming with sub-second latency for interactive voice applications. [4] A second Interactions API, called gemini-3.5-transcribe, is distinctly designed to process pre-recorded audio, meetings, and call logs. [4] This pre-recorded audio processing tool includes built-in features for generating speaker attribution and word-level timestamps. [4]

Gemini 3.5 Transcribe features function calling capabilities that allow the transcription model to delegate complex tasks to other Gemini models. [4] According to Google, examples of these delegable tasks include generating images and analyzing files. [4] This specific function calling capability is currently available within the Gemini macOS application. [4] Ultimately, the transcription system is purposefully designed to capture a user's natural speaking style to better understand their intent and recognize custom vocabulary. [4] The tool is designed to convert raw audio directly into accurate, polished text without struggling with complex jargon. [4]

Google is advancing its speech recognition strategy by baking editing and reasoning capabilities directly into the transcription layer. By offering sub-second latency and function calling, Google positions Gemini 3.5 Transcribe as a powerful endpoint for voice agents and automated workflows. The separation into live and pre-recorded APIs allows developers to optimize for either speed or granular attribution depending on their specific use cases. What the sources don't address: How the model handles complex background noise in real-world enterprise environments compared to conventional speech recognition tools.