Common questions

Voice Inbox runs recognition locally after the speech model is installed. The app needs network access for the first model download, but it is not designed around app-cloud transcription.

Local speech recognition needs a speech model on the device. The pinned int8 Parakeet model download is about 640 MiB, and installation needs about 704 MiB of free storage.

Android targets ARM64 devices running Android 7.0 (API 24) or later. iOS support is currently an MVP and not yet presented as a full store-ready release.

iOS has an active MVP: import audio files, preview playback, speech model download, local transcription, transcript display, and output append are available in the development app. Folder automation, scheduled background transcription, and full Android parity are still in progress.

Android uses Android platform decoders. Tested input includes PCM WAV, AAC in M4A/MP4, MP3, ADTS AAC, mono or stereo FLAC, Ogg Vorbis, Ogg Opus, AMR-NB, and AMR-WB.

iOS uses platform audio decoding in the MVP. Encrypted, DRM-protected, malformed, unsupported codec/container, and unsupported multichannel inputs may be rejected.

It appends new entries to the selected output document. It does not search, replace, or delete previous Markdown/text content, because the document may be edited independently.

No exact timing is promised on Android. Voice Inbox uses Android background work, and Android may delay scheduled work depending on system state. Scheduled background transcription is not part of the current iOS MVP.

A file can fail because decoding, normalization, recognition, staging, or output append publication failed. Failed files appear in the Processed tab with the latest error.

Retry starts a transcription run for one failed file when the app is otherwise ready. Failed and processed files are excluded from Transcribe All.

Voice Inbox identifies files by provider URI and compares reported size and modification time. If known metadata changes, the file becomes pending again.