As with regular chats in Duck.ai, voice chat and dictation are private, anonymized by us, and not used to train AI models. Audio is not stored by DuckDuckGo or by OpenAI (the model provider for voice chat and dictation) after the chat ends, and we do not use audio to identify you. However, your voice is unique to you and can be a biometric identifier. We want you to understand this risk and make an informed choice before using these features.
How voice chat works
When you start a voice chat, Duck.ai connects you to an OpenAI model through an encrypted relay connection that we cannot decrypt. Your microphone audio is streamed in real time to OpenAI, where the model listens, transcribes what you say, and generates both a spoken and text response. Voice chat is currently available in the DuckDuckGo browser and most 3rd-party browsers (Firefox support coming soon). As with text chats, there's a daily limit on voice chats. DuckDuckGo subscribers have access to higher usage limits than the free version of Duck.ai.
How dictation works
When you press the microphone icon in Duck.ai to start dictation, Duck.ai will start recording audio locally, for up to 5 minutes per session. Once you stop the recording, the audio file is sent to OpenAI for transcription via DuckDuckGo's server. OpenAI returns transcribed text to the Duck.ai chat input field, where you can edit it before sending to the model of your choice.
How your voice data is handled
DuckDuckGo protects your audio streams, recordings, and voice data in several ways, to ensure they can only be accessed by OpenAI, (the model provider for voice chat and dictation) and only for the purpose of responding to or transcribing your prompts:
- Ephemeral processing: Your live audio is streamed only while you’re speaking. Once the your voice chat or dictation session ends, neither DuckDuckGo nor OpenAI keeps recordings of your voice.
- No training: Your audio and the model’s responses are not used to train AI models.
- Secure live audio streaming: Voice chat audio streams are encrypted in transit through WebRTC and our relay server.
- Zero data retention: Neither DuckDuckGo nor OpenAI retains any conversation data after you end a voice chat or dictation session.
About your voice and biometric data
Your unaltered voice is unique to you, meaning it can act as a biometric identifier and could theoretically be used to recognize you. DuckDuckGo does not attempt to identify, fingerprint, or link voices to users, and OpenAI is contractually limited in how they can use voice chat and dictation data to what is necessary to perform their services.
You can always opt out of voice chat or dictation entirely. Even when these features are enabled, your audio is only sent to OpenAI when you initiate a voice chat or dictation session, and is deleted after you end it. To opt-out, go to Duck.ai settings and disable the voice chat and dictation toggles. Once you've disabled voice chat or dictation, you'll need to re-enable them in Duck.ai settings before you can use them again.
Please also see our Duck.ai Privacy Policy