Skip to main content

Search & Transcription

The Search tab finds cards across your whole library by what they are called, what is said on them, and how you have marked them. To search what you said in your own recordings, the app first transcribes them — which also lets the review player highlight each word as it is spoken.

Search your cards​

Open the Search tab and start typing. Results match on:

BadgeMatches
TitleThe card's name.
TTSThe text a text-to-speech side was generated from.
ASRThe transcript of a recording (automatic speech recognition).
TagA tag on the card.
FlagThe card's flag.

Each result shows which folder the card is in. Cards in a deactivated folder are marked Deactivated so you know why they are missing from reviews. Tap a result to open the card.

If nothing turns up, try another spelling or fewer words, and remember that recordings only match once they have been transcribed.

How recordings are transcribed​

Text-to-speech sides are searchable immediately, because the app already knows the text. Recordings and imported audio need transcribing first:

  • When you open a card, any side without a transcript is transcribed in the background. You will see Transcribing Audio beside the side while it works.
  • For the whole library, use Settings > Transcription.
  • For one side, tap ⋯ beside the side and choose Transcribe Audio.

Transcription uses each folder's Folder Language, so set that first for anything not in your global language. A clip that was transcribed but contained no words shows No speech detected.

Transcribing audio also records word timings, so the review player can highlight words as they are spoken and let you tap a word to jump to it. See Highlight spoken words.

Transcription​

Settings > Transcription processes every audio file in your library at once.

Choose an engine​

EngineHow it works
Apple SpeechUses the recognition built into your device. Nothing to download. Apple may use online recognition when on-device recognition is not available for a language.
WhisperRuns an OpenAI Whisper model entirely on your device. Choose a Model and download it first; larger models are more accurate but slower and bigger.

Global Language is used for folders without a language of their own. With Whisper, multilingual models detect the language automatically when a folder's language is not supported, while English models always transcribe as English.

Manage Whisper Models lists the models you can download, which are installed, and lets you remove them. Removing a model does not remove transcripts it already made.

Read the summary​

The large number at the top is how many files still need text or cues. Below it:

RowMeans
Audio filesEvery audio side in the library.
TranscribedFiles with a current transcript.
BlankTranscription ran and heard no words — usually silence, very quiet audio, or the wrong language for that side. Tap to see which.
MissingNever transcribed.
ChangedRe-recorded or edited since they were transcribed.
Needs word timingsTranscribed, but without word timings.
TTS audio filesText-to-speech sides, which keep their original text and only need cues.

Tap Start Transcribing to process everything pending. Keep the app open while it runs; Stop Transcribing pauses it, and it picks up where it left off next time. Refresh rescans the library.

Reprocess files​

Invalidation Rules > Force reindexing re-runs transcription on files that already have a transcript:

  • Only missing or changed — the default; nothing extra is reprocessed.
  • All transcribed files — start again from scratch, for example after switching engine.
  • Before date — files transcribed before a date you choose.
  • By model — files transcribed by a particular model, to upgrade them to a better one.

Forced by rule shows how many files the rule adds. If you re-record sides, old transcripts can be left behind; Remove Stale Results clears them without touching any audio.