Voice Inbox

🎙️ Voice Capture & Task Extraction

The Voice feature in Ilseon is designed for high-velocity capture when you are on the move – driving, walking, or in the middle of a workout. It ensures that no task or thought is lost just because you can’t reach your keyboard.

Key Capabilities

Persistent Capture: A dedicated foreground service ensures that recordings aren’t interrupted by the OS, even if you navigate away from the app.
BYOK (Bring Your Own Key): To keep the app lightweight and private, AI features are unlocked by providing your own Gemini API Key in the settings.
Intelligent Task Extraction: Once a recording is transcribed, Ilseon uses Large Language Models (LLMs) to scan the text and automatically extract actionable tasks, separating them from general context.

Open Storage & Data Portability

Ilseon follows a “local-first” philosophy. Your recordings are your own, stored in a transparent directory structure rather than hidden in a private database:

Location: Recordings/ilseon/ on your local device storage.
Format: Industry-standard .m4a files.
Fidelity: Audio is captured at 44.1kHz with a 128kbps bitrate using the AAC_ELD (Enhanced Low Delay) encoder. A professional-grade clarity whith a small storage footprint.
Ownership: Because these files are stored in a standard directory, they can be backed up, shared, or managed using ordinary OS functions or third-party cloud sync tools.

Technical Implementation

The voice workflow is a two-stage pipeline designed for privacy and flexibility:

Local Audio Processing: The app handles the audio recording using a background service to maintain state across different Android lifecycle events.
AI-Powered Refinement:
- Transcription: Converts raw audio into text, saved as a persistent Note.
- Task Distillation: If a Gemini API key is present, the app sends the transcript to the model (gemini-2.5-flash) with a specific system prompt to identify and format “Task” items.