Google has introduced Gemini 3.5 Transcribe, a speech-to-text model designed to handle intelligent voice interactions with improved accuracy compared to conventional speech recognition systems. According to Google, the model addresses challenges like background noise, complex jargon, and disfluency cleanup by converting raw audio directly into accurate, polished, formatted text.
The model is available through two separate APIs. The Live API provides real-time streaming with sub-second latency for interactive voice applications using the gemini-3.5-transcribe-live endpoint. The Interactions API handles pre-recorded audio processing for meetings, call logs, and other stored audio, offering speaker attribution and word-level timestamps.
Gemini 3.5 Transcribe achieves a Word Error Rate (WER) of 4.0% for streaming and 2.6% for non-streaming use-cases, as measured by Artificial Analysis. The model shows strong performance in noisy, real-world environments and accurately captures alphanumeric entities like postal codes and order IDs. Google reports a 70% improvement in time to final transcription compared to its previous model, Chirp 3.
Key features include smart transcription that handles self-corrections, removes filler words, and auto-formats text. The model can delegate complex tasks to other Gemini models via function calling, currently available in the Gemini macOS app. Custom vocabulary support allows the model to recognize specialized jargon and unique spellings. The system automatically detects and transcribes over 85 languages while seamlessly handling regional accents and diverse dialects. For pre-recorded audio, it can attribute speech to up to three speakers, with experimental support for more.
Google has integrated 3.5 Transcribe into several existing products. Rambler on Android transforms spoken thoughts into well-formatted text, while the Gemini app on macOS enables voice commands paired with screen context. The model is available in public preview for developers via Google AI Studio and Google Antigravity, and through Gemini Enterprise Agent Platform for enterprises. It’s already live in the Gemini app on macOS in English, Rambler on Android in select countries and languages, with availability coming soon to Chrome. Developer platforms including Agora, Fishjam, LangChain, LiveKit, Pipecat, Vercel, and Vision Agents are leveraging the Gemini Live API to build voice-driven interfaces.
Key facts
- Gemini 3.5 Transcribe achieves 4.0% word error rate for real-time streaming and 2.6% for pre-recorded audio
- The model supports automatic detection and transcription of over 85 languages with regional accent handling
- Time to final transcription improves by 70% compared to the previous Chirp 3 model
- Multi-speaker identification supports up to three speakers in pre-recorded audio (3+ speakers experimental)
- Available via two APIs: Live API for real-time streaming and Interactions API for pre-recorded audio processing
- Features include smart transcription, custom vocabulary recognition, and function calling capabilities