@@ -39,8 +39,24 @@ If you match these conditions but still don't have access to the feature, reboot
## Privacy and Security
### How does it work?
The Voice to Text feature uses the OpenAI Realtime Transcription API, specifically configured to use the `gpt-4o-mini-transcribe` model. This model is chosen as it supports real-time transcription, which is not available with the `whisper-1` model. This ensures high accuracy and efficiency in converting speech to text. More information about the OpenAI Realtime transcription API available at [https://platform.openai.com/docs/guides/realtime-transcription](https://platform.openai.com/docs/guides/realtime-transcription).
To enhance privacy and security, the Murena Voice to Text keyboard employs a proxy to anonymize requests made to the OpenAI API. This set-up ensures that your data remains secure and your privacy is protected. This is also explained in our [Terms and Conditions](https://murena.io/en/termsandconditions).
### When are data sent to the servers?
Data are sent if and only if:
1. the feature is enabled
2. _AND_ the user had tapped on the mic icon in the keyboard to access the feature.
Outside of this specific case the feature is inactive and no data are sent.
### Does an /e/ OS user’s voice ever get sent to OpenAI’s servers?
If the user is dictating text using Murena Voice to text feature, the audio generated doing so is sent to OpenAI's servers so they can transcribe it into text. As soon as the content has been transcribed the audio file is deleted on the servers.
### Source code
For transparency and further assurance, the source code for the Murena Voice to Text keyboard is available at [https://gitlab.e.foundation/e/os/murena-voice-to-text](https://gitlab.e.foundation/e/os/murena-voice-to-text). Additionally, the source code for the proxy is available at [https://gitlab.e.foundation/e/infra/proxy-stt](https://gitlab.e.foundation/e/infra/proxy-stt).
