Privacy Policy for OpenVox – Local Voice AI

Overview

OpenVox – Local Voice AI (“OpenVox“, “we“, “our“, or “the App“) is committed to protecting your privacy. This Privacy Policy explains our data practices for the OpenVox macOS, iPadOS, and Windows applications.

OpenVox does not provide celebrity voice models and does not permit cloning, impersonating, or commercially using any person's voice without proper rights or consent.

Our Privacy Commitment

OpenVox is privacy-first by design. All text-to-speech and voice processing happens entirely on your device using local machine learning models. OpenVox does not send your prompts, generated audio, or cloned voice audio to third-party AI inference services. Limited network request metadata may be transmitted only when you initiate model downloads from Hugging Face. Optional speech recognition requests during voice cloning (for transcript detection) are handled separately as described in this policy.

Privacy-first architecture does not create permission to use someone else's voice. You are responsible for ensuring you have the necessary rights and consent for any voice samples, cloned voices, generated speech, or commercial use.

Data We Do NOT Collect

OpenVox does NOT collect, use, or transmit the following to OpenVox servers or third-party AI inference services:

• ✗ Text content you generate speech from
• ✗ Generated audio files
• ✗ Voice preferences or settings
• ✗ Usage analytics or telemetry
• ✗ Crash reports or diagnostics
• ✗ Personal content for profiling, advertising, or resale
• ✗ Text prompts/content or generated audio to third-party AI inference services
• ✗ Network activity or browsing data beyond standard model download requests you initiate

Data Storage

All data created by OpenVox remains exclusively on your device:

Local Storage Only

• Generation History: Text inputs, voice selections, and generated audio files (AI Speech and Voice Changer operations) are stored in local application storage. Apple platforms use SwiftData; Windows uses the app's local data storage.
• AudioBook Data: Book titles, chapter information, and generated audiobook files stored locally on your device
• Cloned Voices: Voice embeddings and training data for custom voices stored locally (if voice cloning is used)
• User Preferences: App settings (default voice, speed, temperature, preprocessing/post-processing options) are stored locally using the platform's application-preference storage
• AI Models: Multiple machine learning models are cached locally after initial downloads (OmniVoice, Kokoro-82M, Chatterbox models, Qwen3 TTS models, Voice Conversion models). On macOS, models are stored in ~/.cache/huggingface/. On iPadOS and Windows, models are stored in local app-managed storage.
• Audio Files: Generated speech audio saved to your device's temporary directory and optionally to your chosen save location in WAV, MP3, or M4A formats

Your Control

• • Delete generation history at any time from within the App (History → Delete)
• • Clear specific audiobooks or chapters (AudioBook → Delete)
• • Remove cloned voices (Clone → Manage Voices → Delete)
• • Clear all cached models via Settings → Models (on macOS you may also delete ~/.cache/huggingface/hub/ manually)
• • Uninstalling OpenVox removes app-managed data according to the uninstall and application-data behavior of macOS, iPadOS, or Windows; files exported to locations you selected may remain

Network Activity

OpenVox requires internet access only for initial downloads of AI models from Hugging Face (a public model repository). These downloads:

• • Occur when you first use a feature or manually download a model
• • Download sizes vary by model:
  • - Kokoro-82M: ~350MB
  • - OmniVoice: ~3GB
  • - Chatterbox Turbo: ~400MB-3GB
  • - Chatterbox Multilingual: ~650MB-2.6GB
  • - Qwen3 TTS: ~1.6GB-2.6GB
  • - Qwen3 TTS Voice Design: ~3.1GB
  • - Voice Conversion: ~1GB (for Voice Changer feature)
• • Connect to https://huggingface.co/
• • May transmit standard network request metadata (for example, IP address, user-agent, and related HTTPS request metadata) required for file delivery
• • Are cached locally and never require re-downloading
• • Can be managed via Settings → Models

After initial model downloads, OpenVox functions completely offline (or on local loopback if the Local API is enabled) with no network access required or used for any processing, generation, or voice conversion operations.

Third-Party Model Downloads (Hugging Face)

OpenVox performs text/audio generation locally on your device. Text, prompts, generated audio, and cloned voice audio are not sent to third-party AI services.

To enable local AI features, OpenVox may download model files from Hugging Face, Inc. (huggingface.co) after you grant permission in-app.

When a model download is requested, the following data may be transmitted to Hugging Face

• • Your IP address
• • Standard device/network request metadata (for example, HTTPS request headers such as user-agent and networking metadata)
• • The requested model repository ID and file name

Purpose of this data use

• • To download model files required for local, on-device inference in OpenVox

How this data is used

• • Only for model hosting, delivery, integrity checks, and related download operations

Data sharing

• • OpenVox shares download request data with Hugging Face solely for model file delivery
• • OpenVox does not share your generation input/output content with third-party AI services

User control

• • You are asked for permission before first third-party model download
• • If you do not grant permission, model downloads from Hugging Face will not start

Retention

• • OpenVox does not store your IP address separately for analytics from this flow
• • Downloaded model files are stored locally on your device until you remove them in-app

Third-party protection

• • Third-party processing is limited to providers that apply data protection measures appropriate for secure content delivery and transport encryption (HTTPS/TLS)

Third-Party Services

Hugging Face Model Repository

• Purpose: One-time downloads of TTS and voice conversion models
• Models Downloaded: OmniVoice, Kokoro-82M, Chatterbox models (Turbo, Multilingual), Qwen3 TTS models, Voice Conversion models
• Data Shared: No user-generated content is shared. Standard network metadata (such as IP address and request headers) may be processed by Hugging Face during model downloads, in accordance with their privacy policy.
• Privacy Policy: https://huggingface.co/privacy

Apple Speech Recognition

• Availability: Apple platforms only
• Purpose: Optional transcript detection for uploaded voice-cloning samples
• Data Shared: If you use this feature on a supported Apple platform, the selected uploaded sample or a trimmed excerpt, plus language/recognition request metadata, may be processed by Apple to return a transcript for your review
• Privacy Policy: https://www.apple.com/legal/privacy/

Privacy Policy Metadata (Model Downloads)

• Data Category: Identifiers / network metadata (IP, request metadata)
• Purpose: App functionality (model download)
• Third Party: Hugging Face, Inc.
• Collection Method: During user-initiated model download requests

OpenVox does not integrate with any analytics services, advertising networks, or data brokers.

Permissions Required

File System Access

• Purpose: Save generated audio files and import source files (PDF, text, audio)
• Scope: Only locations you explicitly select via the platform file picker
• Usage:
  • - Write audio files (WAV, MP3, M4A formats)
  • - Read PDF/text files for AudioBook generation
  • - Read audio files for Voice Changer input
• Your Control: File access granted per-operation via the standard platform file picker

Microphone Access

• Purpose: Voice cloning from live audio recording
• Scope: Recording limited to active cloning sessions when you choose “Record” option
• Usage: Audio captured for voice training, processed locally, never transmitted
• Alternative: Upload pre-recorded audio files instead of using microphone

Speech Recognition Access

• Purpose: Detect the spoken reference script from uploaded audio during Voice Cloning
• Scope: Available on supported Apple platforms and used only if you choose automatic transcript detection and grant speech recognition permission
• Usage: May send the selected uploaded sample, or a trimmed excerpt, to Apple's Speech Recognition service to return a transcript for your review
• Alternative: You can deny this permission and enter the reference transcript manually

No Other Permissions

OpenVox does NOT request:

• • Camera access
• • Location access
• • Contacts access
• • Calendar access
• • Photo library access

Local API Integration

Note: The Local API is a macOS-only feature and is not available on iPadOS or Windows.

OpenVox for macOS includes an optional Local API that allows other applications on your Mac to interact with OpenVox for speech generation and model management.

• Loopback by Default: The Local API is hosted on your Mac at http://127.0.0.1:8000. By default, it is only accessible to applications running on the same machine.
• Local Network Access: If you intentionally configure the API for local network access, it may be accessible to other devices on your same Wi-Fi or wired network. This remains within your controlled environment and is never exposed to the public internet by OpenVox.
• No External Communication: The Local API does not communicate with OpenVox servers or any third-party AI services. All requests and audio generation remain local to your hardware.

Children's Privacy

OpenVox does not knowingly collect or solicit generation content from children under 13. OpenVox does not send generation content to third-party AI services, and only limited network metadata may be processed during user-initiated model downloads.

Platform Acceleration & On-Device Processing

OpenVox uses Apple's MLX framework via Python for hardware-accelerated AI processing on macOS and Core ML / ANE (Apple Neural Engine) on iPadOS. On Windows, supported models run locally using the compute resources available on the PC, including supported GPU acceleration where available. OpenVox itself does not intentionally send generation data to Apple, Microsoft, or any external service during processing. Optional Apple speech-recognition requests during voice cloning are handled separately as described in this policy.

System-Level Data Disclaimer

While OpenVox itself does not collect analytics or diagnostics, macOS, iPadOS, and Windows may generate system-level logs or crash reports according to your system settings. These reports are controlled entirely by Apple (macOS/iPadOS), Microsoft (Windows), and the user, and are not accessible to OpenVox.

Changes to This Policy

We may update this Privacy Policy to reflect changes in the App or legal requirements. Updates will be posted with a new “Last Updated” date. Continued use of OpenVox after changes constitutes acceptance of the updated policy.

Data Protection Rights

OpenVox keeps generation content local on your device, and OpenVox does not maintain user content databases. You maintain direct control over data created by the App through in-app deletion options and standard platform file management. Limited model-download request metadata may be processed by Hugging Face as described in this policy.

Your Rights

• Right to Privacy: All processing is local by design
• Right to Delete: Remove generation history or uninstall the App at any time
• Right to Export: Copy generated audio files to any location
• Right to Transparency: Full source code and architecture documented

Data Retention

OpenVox does not retain generation content on OpenVox-controlled servers. All locally stored data persists only until you delete it or uninstall the app. Model-download request metadata is handled by Hugging Face according to its own retention practices.

Contact Us

For privacy questions or concerns:

• Email: [email protected]
• Website: https://theoracleguy.in

Legal Basis (GDPR)

For users in the European Economic Area: OpenVox relies on performance of a contract (providing requested app functionality) and user consent (for user-initiated third-party model downloads and optional speech recognition). Generation content is processed locally on-device. Limited identifiers/network metadata may be processed during model downloads by Hugging Face as an infrastructure provider. Optional speech recognition data is processed by Apple only upon user consent.

California Privacy Rights (CCPA)

For California residents: OpenVox does not “sell” personal information and does not share personal information for cross-context behavioral advertising. Limited identifiers/network metadata may be processed by Hugging Face only to deliver user-initiated model downloads. Optional speech recognition requests are processed by Apple only upon user initiation.