Google Gemini 3.5 Transcribe: Speech-to-Text for Gboard and Chrome

Google has significantly expanded its speech AI portfolio with the introduction of Gemini 3.5 Transcribe, a next-generation speech-to-text model that converts raw audio directly into accurate, polished, and formatted text. The model, which builds upon the capabilities first demonstrated in the Gboard Rambler feature announced at The Android Show I/O Edition in May 2026, is now rolling out across multiple Google products including the Chrome browser, Gemini macOS app, and enterprise platforms.

Unlike conventional speech recognition systems that struggle with background noise, technical jargon, and conversational disfluencies, Gemini 3.5 Transcribe is designed to understand natural speaking styles, recognize custom vocabulary, and deliver publication-ready text with minimal post-editing.

From Gboard Rambler to Gemini 3.5 Transcribe

The foundation for Gemini 3.5 Transcribe was laid in May 2026 when Google announced Rambler, an AI-powered voice dictation feature for Gboard — the default keyboard app for hundreds of millions of Android devices worldwide. Rambler was positioned as a core component of Google's broader Gemini Intelligence suite for Android.

"Whether you're blending English with Hindi or any other combination, Rambler understands the context and the nuance, ensuring your message sounds exactly like you — only more polished."

— Google, The Android Show I/O Edition, May 2026

Rambler introduced several key innovations that are now central to Gemini 3.5 Transcribe:

• Filler word removal: Automatically eliminates "ums," "ahs," and other speech disfluencies for cleaner output
• Self-correction understanding: Recognizes when users verbally correct themselves mid-sentence and outputs only the final intended text
• Code-switching support: Seamlessly handles multilingual speech where users switch between languages mid-sentence — a capability reflecting how many multilingual speakers actually communicate
• Privacy-first processing: Audio is used only for real-time transcription and is never stored or saved

The initial Rambler rollout was limited to Samsung Galaxy and Google Pixel devices, with broader Android expansion planned for later in 2026. Gemini 3.5 Transcribe represents the generalization and API-ification of these capabilities, making them available to developers and enterprises beyond Google's first-party apps.

Technical Capabilities and Benchmarks

Gemini 3.5 Transcribe delivers measurable improvements over its predecessor, the Chirp 3 transcription model from 2025. Independent benchmarking by Artificial Analysis confirms significant advances in both accuracy and speed:

Metric	Gemini 3.5 Transcribe	Previous Generation (Chirp 3)
Streaming WER	4.0%	~6.5%
Non-Streaming WER	2.6%	~4.2%
Time to Final Transcription	70% faster	Baseline
FLEURS Benchmark (Streaming)	5.50% WER	~7.8% WER
FLEURS Benchmark (Non-Streaming)	5.04% WER	~7.2% WER

The Word Error Rate (WER) improvements are particularly significant for real-world applications. A 2.6% non-streaming WER means that out of every 1,000 words transcribed, only approximately 26 require correction — a threshold that makes automated transcription viable for professional use cases including journalism, medical documentation, and legal proceedings.

Key Features

Custom Vocabulary: The model recognizes specialized jargon, brand names, medical terminology, and unique spellings by adapting to user-provided custom vocabulary lists. This is critical for industries where generic speech recognition fails on domain-specific terms.

Global Language Support: Gemini 3.5 Transcribe automatically detects and transcribes over 85 languages, handling regional accents and diverse dialects without manual language selection. This exceeds the 70+ language support of the related Gemini 3.5 Live Translate model.

Multi-Speaker Identification: For pre-recorded audio, the model accurately attributes speech to individual speakers with timestamps for up to three speakers. Experimental support for more than three speakers is available for enterprise use cases.

Noise Robustness: The model maintains accuracy in noisy, real-world environments and accurately captures alphanumeric entities such as postal codes, order IDs, and phone numbers — a common failure point for earlier speech recognition systems.

Product Integration and Availability

Google is deploying Gemini 3.5 Transcribe across its ecosystem through three primary channels:

Consumer Products

• Gboard Rambler (Android): The flagship consumer experience, enabling polished dictation in any app that supports Gboard. Users can dictate messages, emails, notes, and social media posts with automatic cleanup of filler words and self-corrections
• Gemini macOS App: Desktop transcription for Mac users, enabling voice input for prompts, document drafting, and creative writing
• Chrome Browser (Coming Soon): Integration into Chrome will allow users to "talk to type" in any web field — dictating replies, drafting posts, or prompting Gemini directly through voice without touching the keyboard

Developer Access

Developers can access Gemini 3.5 Transcribe in public preview through:

• Gemini API: Standard REST API integration for batch and streaming transcription
• Google AI Studio: Browser-based testing and prototyping environment
• Google Antigravity: Advanced prompt box microphone that pairs screen context and chat history with transcription for pinpoint accuracy across file names, agent thoughts, and active documents

Enterprise Platforms

For business customers, Gemini 3.5 Transcribe is available in public preview via:

• Gemini Enterprise Agent Platform: For building custom voice-enabled business applications
• Gemini Enterprise for Customer Experience (Coming Soon): Will enable real-time transcription of customer service calls, automated note-taking, and sentiment analysis

Competitive Positioning

The speech-to-text market has become increasingly crowded with specialized startups and platform providers. Google's entry with Gemini 3.5 Transcribe leverages several strategic advantages:

Competitor	Key Strength	Google's Advantage
OpenAI Whisper / GPT-4o-transcribe	High accuracy; strong open-source ecosystem	Native integration with Android, Chrome, and Workspace; no separate API needed
Deepgram	Developer-friendly; real-time streaming	Pre-installed on billions of Android devices via Gboard
AssemblyAI	Enterprise features; speaker diarization	Multilingual code-switching; custom vocabulary at consumer scale
Eleven Labs	Voice cloning; high-quality TTS	End-to-end ecosystem from transcription to translation to synthesis

Industry benchmarks from early 2026 placed GPT-4o-transcribe at the top of speech-to-text accuracy rankings, followed by Eleven Labs and Whisper-large, with Gemini models in the top tier. Gemini 3.5 Transcribe's improvements may close or reverse this gap, particularly for multilingual and noisy-environment use cases.

Gemini Enterprise for Legal

Alongside the transcription model rollout, Google Cloud launched Gemini Enterprise for Legal on August 25, 2026 — a purpose-built version of its enterprise platform configured for law firms and corporate legal departments. The platform leverages Gemini 3.5 Transcribe for:

• Real-time transcription of depositions and court proceedings
• Automated generation of legal documents from voice dictation
• Analysis of audio evidence with speaker identification and timestamping

Launch partners include Cleary Gottlieb, Freshfields, Weil, and Williams & Connolly. The product is currently available in preview and represents Google's push to dominate professional verticals where transcription accuracy directly impacts billable hours and case outcomes.

Privacy and Data Handling

Google has emphasized privacy as a core differentiator for Gemini 3.5 Transcribe. According to Ben Greenwood, Director of Android Core Experiences, Google uses a combination of on-device and cloud-based processing and has "invested significantly over many years" to ensure features are safe and private.

Key privacy commitments include:

• No audio storage: Voice recordings are used only for real-time transcription and are not retained
• Clear user indication: Gboard clearly indicates when Rambler is active, ensuring users know when voice processing is occurring
• On-device options: For sensitive use cases, future updates may enable fully on-device transcription without cloud transmission, similar to the offline-first Google AI Edge Eloquent app released for iOS in April 2026

Conclusion

Gemini 3.5 Transcribe represents Google's most serious bid yet to own the speech-to-text layer of the AI stack. By combining industry-leading accuracy (2.6% WER), multilingual code-switching, custom vocabulary support, and deep integration across Android, Chrome, and enterprise platforms, Google is leveraging its massive distribution advantage to make polished voice dictation the default input method for billions of users.

For consumers, the promise is a keyboard that truly understands how people speak — messy, multilingual, self-correcting, and filled with "ums" — and turns it into clean, professional text. For developers and enterprises, the API and platform integrations offer a path to build voice-enabled applications without managing complex speech infrastructure. And for competitors in the dictation and transcription space, Google's entry at the operating-system level means the battle for market share just became significantly harder.

As Chrome integration rolls out and enterprise features mature through 2026, Gemini 3.5 Transcribe is positioned to become as ubiquitous for voice input as Google Search is for text queries.

Frequently Asked Questions (FAQ)

• Q1: What is Gemini 3.5 Transcribe and how accurate is it?

  Gemini 3.5 Transcribe is Google's advanced speech-to-text model that converts raw audio into polished, formatted text. It achieves a 2.6% Word Error Rate for non-streaming use and 4.0% for streaming, with 70% faster time-to-final-transcription compared to the previous Chirp 3 model.

• Q2: Where can I use Gemini 3.5 Transcribe?

  The model is available in Gboard Rambler on Android, the Gemini macOS app, and Google Antigravity. It is coming soon to the Chrome browser. Developers can access it via the Gemini API and Google AI Studio, while enterprise customers can use it through the Gemini Enterprise Agent Platform.

• Q3: What languages does Gemini 3.5 Transcribe support?

  Gemini 3.5 Transcribe automatically detects and transcribes over 85 languages, including support for regional accents, diverse dialects, and mid-sentence code-switching between languages.