EchoNote: Audio Transcription

EchoNote records speech and turns saved audio into editable text on compatible Android devices.

Private, on-device transcription
Audio, recording titles, and transcript text are processed and stored in the app's private storage. EchoNote does not upload audio, titles, or transcript text. No account is required. After a verified model is installed, recording, transcription, playback, editing, search, copy, and export can work offline.

What you can do
- Record, pause, resume, and save audio
- Import supported audio or MP4 video
- Transcribe saved recordings on the device
- Choose a preferred recognition language or automatic detection
- Play the original audio and edit the transcript
- Search, copy, share, or export TXT, Markdown, SRT, and WebVTT files
- Organize recordings with folders, favorites, and recently deleted items
- Add custom vocabulary for names, brands, and specialist terms

Models, network, and diagnostics
Speech models are optional downloads from 30.7 MiB to 547.4 MiB. Small q5_1 (181.3 MiB) is the recommended balanced model in current testing, and the shared voice-activity file is about 0.9 MiB. The app shows each size and verifies every file before use. Model download hosts receive network metadata such as the IP address, requested model path, Range header, and request time, but not recorded audio, titles, or transcript text.

Release builds use Firebase Analytics and Crashlytics for app-usage statistics, crash logs, diagnostics, and device or installation identifiers. The advertising ID is disabled. EchoNote does not place audio, recording titles, transcript text, custom vocabulary, or imported media in analytics or crash reports.

Storage and compatibility
Export happens only after you choose a destination through the Android system picker. Uninstalling EchoNote permanently deletes its private recordings, transcripts, settings, vocabulary, and downloaded models from the device, so export important content first.

EchoNote requires Android 8.0 or newer on an arm64-v8a device. Recognition quality and processing time vary by language, audio, model, device memory, and processor. Long recordings and larger experimental models can take significant time or fail on devices with limited resources.