Google Announces Gemini 3.5 Transcribe for Smarter AI-Powered Speech-to-Text

Google has introduced Gemini 3.5 Transcribe, a new speech-to-text model designed to turn spoken audio into accurate, polished and intelligently formatted text.

Announced on August 26, 2026, the new model is part of Google’s Gemini Audio technology and is designed to go beyond traditional transcription. Instead of simply recording every word exactly as spoken, Gemini 3.5 Transcribe can understand context, remove unnecessary filler words, recognize specialized terminology and format the resulting text automatically.

What Makes Gemini 3.5 Transcribe Different?

Traditional speech-recognition systems generally attempt to reproduce audio as literally as possible. Google’s new model takes a more intelligent approach.

Gemini 3.5 Transcribe can remove verbal fillers such as “um” and “uh,” clean up repeated phrases and handle situations where speakers correct themselves while talking. The result is intended to look more like a finished piece of writing rather than a raw transcript.

Google says the model is particularly designed for difficult audio conditions, including background noise, accents, technical vocabulary and conversations involving multiple languages.

The model automatically detects more than 85 languages and can handle language switching during a conversation. Google also supports custom vocabulary, allowing developers to provide specialized terms, names, abbreviations or product terminology that the system should recognize accurately.

Real-Time Transcription Is a Major Focus

Google is offering Gemini 3.5 Transcribe in two versions.

The standard gemini-3.5-transcribe model is designed for recorded audio, such as meetings, interviews, podcasts and customer-service calls. It supports speaker identification and word-level timestamps.

The gemini-3.5-transcribe-live version is built for real-time applications. It can stream audio through Google’s Live API and is aimed at voice agents, live captions and other interactive experiences.

Google says the new model delivers significantly lower transcription latency than its previous Chirp 3 system, making it more suitable for applications where users expect an immediate response.

Speaker Identification and Custom Vocabulary

For recorded audio, Gemini 3.5 Transcribe can distinguish between speakers and assign sections of a conversation to different voices. The API documentation says speaker diarization can support up to eight speakers, although attribution involving more than three speakers remains experimental.

Developers can also supply up to 1,000 custom vocabulary terms. This could be particularly useful for industries that rely on product codes, medical terminology, company names, technical language or other words that conventional speech recognition systems may struggle to identify.

Google Is Bringing the Technology to Its Products

The technology is already being used across some Google products and features.

Google says Gemini 3.5 Transcribe is powering new voice capabilities including Rambler, a dictation feature on Android, as well as voice experiences in the Gemini app for macOS. Developers can access the model through the Gemini API in Google AI Studio and the Gemini Enterprise Agent Platform.

Google has also indicated that broader Chrome integration is coming, potentially bringing the improved transcription technology to more voice-typing and browser-based applications.

What It Could Mean for AI Voice Applications

The launch could make high-quality transcription an increasingly important component of AI assistants and voice agents.

Businesses could use Gemini 3.5 Transcribe for meeting records, call-center analysis, interviews and automated documentation. Developers could also use the real-time version to build voice assistants that can understand users more quickly and respond without waiting for a complete recording to finish.

Google’s API documentation lists Bengali, including Bangladesh and India variants, among the supported languages, expanding the potential usefulness of the technology for multilingual applications.

Gemini’s AI Audio Push Continues

Gemini 3.5 Transcribe represents another step in Google’s effort to make its AI models more capable with spoken language.

Rather than treating speech-to-text as a simple conversion task, Google is positioning transcription as an intelligent part of the AI workflow—one that can understand context, organize information and prepare spoken content for downstream applications.

As developers increasingly build AI agents that communicate through voice, Gemini 3.5 Transcribe could become an important piece of Google’s broader strategy to compete in the rapidly expanding voice-AI market.