Talk Live

Why It Exists

Travelers, multilingual families, and international teams need real-time translation during conversations without privacy concerns or internet dependency.

How It Works

Talk Live is Google's on-device real-time conversation translation feature powered by Gemini Nano. It listens to spoken language, translates it instantly, and outputs the translation as speech or text. The system supports multiple language pairs and works without internet connection on supported Pixel devices. All speech recognition and translation processing happens locally using the Tensor NPU, ensuring conversation privacy. Talk Live was introduced with the Pixel 9 series and supports 20+ language pairs.

Everyday Use Cases

• International travel
• Multilingual family conversations
• Business meetings
• Language learning
• Healthcare appointments
• Academic conferences

User Workflow

• User opens Talk Live and selects language pair
• Both participants speak in their native languages
• Gemini Nano transcribes and translates in real-time
• Translation is played as speech or displayed as text
• Conversation continues with seamless translation

AI Processing Flow

Speech audio is captured by the device microphone and processed by on-device Gemini Nano speech recognition models. The recognized text is translated by an on-device neural machine translation model. The translated text is then synthesized into speech using a neural text-to-speech model. All processing uses the Tensor NPU. Multiple language pairs are supported with on-device models.

Inputs / Outputs

Inputs:

• Spoken language audio
• Selected language pair
• Ambient noise profile
• User voice characteristics
• Conversation context

Outputs:

• Translated speech audio
• Real-time text translation
• Bilingual conversation view
• Confidence indicators
• Speaker separation
• Language detection

Known Limitations

• Requires Pixel 9 or newer with Tensor G4
• Supports 20 language pairs at launch
• Translation quality varies by language pair
• Background noise may reduce accuracy
• Conversation must be relatively quiet for best results

Unsupported Scenarios

• Silent text-based translation
• Sign language interpretation
• Real-time group translation with 5+ participants
• Whispered speech
• Extremely noisy environments
• Non-Pixel devices

Performance Notes

Real-time translation with under 300ms latency; Speech recognition in under 100ms; NPU processing for all translation; Battery consumption under 3 percent per hour of active translation

Available On

Shows where this feature is available and how its AI processing works on each platform. Availability may vary by device.

Research Status

Confidence and verification reflect how complete documentation is. Fields may show Not assessed or Not yet verified while research is ongoing - this flags gaps, not product deficiencies.

Research Notes

Available on Pixel 9 and newer with Tensor G4; Introduced October 2024 with Pixel 9; Uses Gemini Nano 2 translation model; 20+ language pairs; End-to-end encryption for conversations; Offline mode supported

Details

Capability Mapping

This feature maps to canonical capability family: Text Summarization.

Canonical capability: Text Summarization.

• Text Summarization - Condenses text, messages, documents, or conversations into a shorter summary.

Grounded in 4 device(s) on Gemini Nano.

Related Features

• Gemini Live - Cross-platform equivalent