Polished Recognition

Polished Recognition

Voice input for any keyboard — transcribe with your own API endpoints

Georg Graf System

Download (1.4 MB)

Description

Polished Recognition is an Android voice input service that works with any keyboard — Gboard, AnySoftKeyboard, Samsung Keyboard, and more.

Instead of relying on a fixed speech provider, you configure your own OpenAI-compatible endpoints:

• STT (Speech-to-Text): Point to any Whisper-compatible API (Groq, OpenAI, self-hosted, etc.)
• LLM (Language Model): Optionally post-process transcriptions through any chat-compatible model
• Custom prompts: Edit system, user, and translation prompts directly in Settings
• Privacy-first: No analytics, no tracking, no proprietary SDKs. All data flows directly between your device and your chosen endpoints.

Simply select "Polished Recognition" as your system's voice input provider, tap the microphone on any keyboard, and start speaking.

Permissions:
• RECORD_AUDIO — required to capture microphone input
• INTERNET — required to send audio to your configured STT/LLM endpoints

Versions

Version Date Size Requires Android Download
1.2.1 current
e252e31acc0795f5…
 2026-09-09 1.4 MB API 30

Permissions

What the app asks the system for. Android only grants sensitive access after you confirm it.

  • MicrophoneRecord audio.
  • Internet accessSend and receive data over the internet.
  • Network stateCheck whether and how the device is connected.
  • Foreground serviceKeep running in the background, visible in the status bar.
  • FOREGROUND_SERVICE_MICROPHONE
  • NotificationsShow notifications.
  • polishedrecognition.DYNAMIC_RECEIVER_NOT_EXPORTED_PERMISSION

Comments