Connect uses cloud AI providers for real-time voice translation. We document the data flows, storage choices and user controls so you can evaluate Connect before using it for personal or business calls.
Data processing
Live audio is streamed only to the providers required by the enabled feature. Soniox is used for real-time speech recognition and translation; Cartesia is used for speech synthesis and enabled voice features. Connect's main workflow does not intentionally create a conversation recording, but provider retention may differ.
- Optional transcripts remain on your device until you delete them.
- Speaker matching may create a temporary Firebase Storage object; Connect attempts to delete it after the request.
- Audio is encrypted in transit.
Deletion and backups
When supported data is deleted through Connect, the active Firestore or Cloud Storage copy is deleted immediately after the deletion operation completes.
- Production Firestore shown in the project console: region
nam5.
- Point-in-time recovery: 7 days.
- Daily scheduled backups: 98 days.
- Backups are separate from active data; a restore or export can create another copy.
What Firestore may contain
Depending on the features and account activity, Firestore may contain operational records such as:
- Accounts and profiles (
users), customers, organizations, plans, subscriptions and licences.
- Voice metadata and optional voiceprints, translation contexts (
soniox_contexts), dictionaries and dataset metadata.
- Feedback, issue reports, activation nudges, notifications, products, credit or usage alerts and newsletter records.
Collection names and fields may change as the product evolves. The active records can be deleted from Connect; retention in provider backups follows the backup lifecycle above.
Voice and sensitive features
Speaker identification, voiceprint enrolment and automated gender estimation are optional. PyannoteAI may process a short sample for speaker identification or enrolment, and Gender Recognition may process a short sample for an automated estimate.
- A voiceprint may be biometric data when used to identify a person.
- Do not submit another person's voice without giving the required notice and obtaining any required permission.
- Provider retention and training terms must be checked in the applicable provider agreement.
Security boundaries
Controls observed in the reviewed application include TLS in transit, Firebase authentication, server-side authorization and provider-managed access controls. Connect is not end-to-end encrypted because enabled cloud providers process the audio.
- Sentry may receive account identifiers and diagnostic metadata.
- Google Analytics and Amplitude are enabled after web consent; Amplitude session replay requires ongoing masking review.
- This page does not claim GDPR, CCPA or APPI certification, ISO 27001 certification, a security audit or a universal uptime SLA.
Privacy rights and subprocessors
Connect's observable provider chain includes Firebase/Google Cloud, Soniox, Cartesia, Sentry, Stripe, Resend, Google Analytics and Amplitude, with PyannoteAI and Gender Recognition used only when their optional features are enabled. Vercel and Hetzner depend on the active deployment configuration.
View the current subprocessor list for each provider's documented role. Hume AI, OpenAI, Anthropic, ElevenLabs and Loops are not listed because their use was not observed in the current production path.
Review the applicable rights and request deletion through the Privacy Policy, or contact privacy@belora-connect.com for privacy and security questions.