Last updated: September 2026 (foreground-only microphone use clarified)
EchoSight is a real-time voice transcription app designed for deaf and hard-of-hearing users. It turns speech into readable text on your screen in real time, detects environmental sounds (doorbells, alarms, etc.) and vibrates to alert you, and supports face-to-face conversation mode. No account is required.
All speech recognition and sound detection happens locally on your device. EchoSight uses on-device AI models (Sherpa-ONNX Zipformer for streaming recognition, SenseVoice for offline refinement, and Zipformer Audio Tagging for environmental sound classification). Microphone audio and transcript text are not transmitted to EchoSight or any third party.
EchoSight uses the internet only to download AI models when you choose to use them for the first time. Models are downloaded from Hugging Face or the hf-mirror.com mirror. These providers receive ordinary network request information, such as your IP address and HTTP request metadata, under their respective privacy policies. EchoSight does not include microphone audio or transcript text in these requests. Once the required models are downloaded, transcription works offline.
EchoSight requests the following permissions:
EchoSight does not request access to:
EchoSight accesses the microphone only after you tap the listening control while the app is visible. Listening stops when you tap stop or when the app moves to the background. EchoSight does not use a microphone foreground service and does not continue recording in the background.
We do not collect personal data on our systems. EchoSight does not transmit to us:
The following data is stored locally on your device only and never leaves your device:
You can delete individual conversation records in the app. App-local data is removed when you uninstall the app, subject to your device's operating-system backup and restore settings.
EchoSight uses Hugging Face and hf-mirror.com only to deliver model files. It does not integrate third-party advertising, analytics, login, or crash-reporting SDKs. Specifically:
The AI models downloadable in EchoSight are open-source: Sherpa-ONNX Zipformer (Apache 2.0), Kroko Zipformer for French/Spanish/German (CC-BY-SA-4.0), SenseVoice (MIT), Qwen3-ASR (Apache 2.0), NVIDIA NeMo Parakeet (CC-BY-4.0), and Zipformer Audio Tagging (Apache 2.0). These models run entirely on your device.
EchoSight is not specifically directed at children under 13. We do not knowingly collect personal data from children.
We do not retain user data on our servers. Data generated by the app, including conversation history and preferences, exists locally on your device until you delete it or uninstall the app, subject to your device's backup and restore settings.
Since EchoSight does not collect, store, or process your app content on our servers:
We may update this privacy policy from time to time. Any changes will be reflected in the "Last updated" date above. Your continued use of the app after changes constitutes acceptance of the updated policy.
If you have questions about this privacy policy, please contact us at:
support@3vonline.com