Gemini Audio
AI transcription
Go beyond simple transcription, identify who’s talking, and understand the intent behind the words
Models
Get clear and helpful insights directly from audio files. Identify speakers, understand the key points they’ve made – and grasp the sentiment behind those points.
3.5 Transcribe
Best for speech transcription. Transcribes pre-recorded audio across 85+ languages with high alphanumeric accuracy and timestamps for up to three speakers.
Precise audio analysis
Unlock insights directly from audio files with Gemini’s audio capabilities.
Slide 1 of 7
Real-time streaming transcription
Transcribe live audio streams, including custom vocabulary, with sub-second latency ideal for building interactive voice agents, real-time captioning pipelines, and live translation layers.
Prerecorded file transcription
Transform prerecorded audio – like voice notes, support calls, or lectures – into clean, structured text. Optimized for precision, custom vocabulary, and speaker attribution.
Clear and actionable data
Transform unstructured audio – like voice notes, support calls, or lectures – into clean and actionable notes, while filtering out filler words. Export as JSON format, in a summary, or as bullet points.
Precise speaker identification
Distinguish and label speakers within a single transcript. Provides clear and correct attribution to support compliance audits and sentiment analysis, or to create accurate meeting summaries.
Alphanumeric accuracy
Recognizes alphanumeric data like phone numbers, postal codes and order IDs with precision – critical in handling customer accounts and transactions.
Broad language support
Automatically detects and transcribes over 85 languages, smoothly navigating regional accents and local dialects across major languages like English, Spanish, French, Turkish, Mandarin, and Japanese
Context recognition
Can be taught to recognize specific vocabulary like brand names, product titles, and industry terminology – to prevent misspelling and misunderstanding.
Advanced AI transcription capabilities
A unified voice experience that cleans up speech, understands intent, and executes tasks.
Disfluency clean-up
Filters pauses, “ums”, “ahs” and other filler words, to produce polished text with accurate punctuation and useful formatting – at the speed of speech.
Function calling
Delegates complex tasks to other Gemini models—enabling image generation, live search, text summarization, and file analysis.
Adaptable voice editing
Refine your thoughts in the moment – correcting details, clarifying spellings, or change the writing style without missing a beat.
Showcase
Slide 1 of 6
“The transcription model is not only fast, but impressively accurate with emails and numbers, two areas where the industry still needs meaningful advancement.”
Mason Adams
Developer Evangelist, Agora
“By integrating Gemini 3.5 Transcribe model, IntelliTek Health transforms real-time clinical transcription across primary care and medical specialties. Gemini’s accuracy and multi-lingual capabilities capture every critical detail, while multi-region availability ensures global data compliance. This integration reduces documentation time, boosting efficiency and empowering healthcare providers to focus on what matters most: developing accurate and highly personalized patient care plans.”
Martyn Molnar
CEO, IntelliTek Health
“Customers trust Lingopal to deliver real-time translation to billions of fans watching broadcasts worldwide. We combine first- and third-party models for the highest quality in every language pair, and Gemini is consistently among our top selections. We're excited to integrate Gemini 3.5 Transcribe into our speech recognition layer, automatically detecting speaker languages and routing multilingual broadcast audio faster than ever.”
Fanley Tseng
Head of Engineering, Lingopal.ai
“We know how hard real-time multimedia processing is. We've been building it for years, so we don't take latency claims at face value. We measured Gemini 3.5 Transcribe and the numbers matched what we were promised!”
Maciej Rys
VP of Engineering, Software Mansion
“The DeepMind team achieved the jump from technical transcription to natural understanding. From native alphanumeric conversion to seamless language identification and switching, they somehow enabled this while keeping the latency sensationally low, making this one of the fastest, most capable transcription options out there.”
Stefan Blos
Senior Developer Advocate, Stream
“At Vivo, we're committed to building AI experiences that break language barriers for users around the world. Leveraging Google's Gemini 3.5 Transcribe enables us to deliver fast, accurate multilingual transcription and translation — turning speech into text people can understand in their own language. It's a truly meaningful step toward borderless communication.”
Chen Min
Senior Director of AI Engineering, Vivo
Performance
Gemini 3.5 Transcribe’s performance represents a major advancement from our previous transcription model, Chirp.
Slide 1 of 2
FLEURS (top locales) Streaming Index
On the FLEURS benchmark across a set of top languages and locales, the model delivers highly precise multilingual performance, achieving a 5.50% WER in streaming mode.
FLEURS (top locales) Non-Streaming Index
On the FLEURS benchmark across a set of top languages and locales, the model delivers highly precise multilingual performance, achieving a 5.04% WER in non-streaming use-cases.
Model information
- Name
- 3.5 Transcribe
- Status
- Preview
- Input
- Text
- Audio
- Output
- Text
- Input tokens
- 96k
- Output tokens
- 32k
- Availability
- Google Antigravity
- Gboard Rambler
- Gemini API
- Gemini App
- Gemini App on MacOS
- Google AI Studio
- Google Enterprise Agent Platform
- Google Workspace
- Documentation
- View developer docs
- Model card
- View model card
Try AI transcription
Google AI Studio
The fastest path from prompt to production
Gemini Enterprise Agent Platform
Build, scale, and govern agents
Gemini API
Get started building with cutting-edge AI models
Gemini for macOS
Supercharge your creativity and productivity
Google Antigravity
Our AI-first development platform that allows anyone to be a builder
Google Workspace
Collaborate, create, and communicate all in one place