Psycholog — user guide
Review every transcript and speaker label against the recording before using it.
Before the first session
Psycholog records conversations, keeps patient records, plays sessions and creates a text transcript after a session. Workflow: obtain consent → New session → record → Stop and save → Session history → Create transcript → optionally Separate speakers → review and correct.
- Use a device passcode; in Settings you may enable the optional app lock with Face ID, Touch ID or the device passcode.
- Choose the recording language: Polish, English, German, Italian, French or Spanish. The interface language is chosen separately.
- Prepare storage: about 200 MiB is required before recording, plus space for recordings, models and exports. One hour uses about 29 MB, 43 MB or 58 MB depending on quality.
- Microphone access is required for sessions and voice samples. If refused, enable it in iOS Settings. You choose file and sharing access in the system picker.
Recording, pause and interruptions
From the home screen select New session, select a patient or enter a name/identifier, confirm that consent was obtained, then tap Start. Check Recording status, elapsed time, microphone level and audio source. Finish with Stop → Finish and save; do not close the app instead of saving.
- Pause stops capturing without ending the session; Resume continues the same session in further segments. Audio during the pause is not recorded.
- Recording pauses when iOS interrupts audio, for example for a phone call or system event. It also pauses when the previous audio device becomes unavailable, for example when the active headset disconnects. Not every route change pauses it.
- A pause can also occur at critically low space (about 20 MiB) or if the next protective segment cannot be created. Remove the cause, return to the app, check the audio source and tap Resume yourself. It never resumes automatically.
- Background recording is enabled, but the lock screen, background operation and iOS cannot guarantee continuity. Check the recording indicators after every interruption.
Transcription, Internet and models
In Session history open a saved recording and choose Create transcript. Text is not made live. First use downloads and prepares missing models, so it needs a stable Internet connection, preferably Wi‑Fi.
- Local transcription uses FluidAudio Parakeet TDT 0.6B v3 INT8 (about 480–485 MB). Speaker separation downloads a separate set of about 22 MB. Creating a voice profile downloads separate resources of about 14 MB. The total is about 520 MB plus working space.
- After the complete set for a function is downloaded, transcription, diarization and voice comparison run locally offline. A missing or damaged model may need downloading again.
- Models are downloaded from FluidInference repositories on Hugging Face. The download is model files, not recordings or patient data. The App Store handles price, purchase and restore.
Speakers, voice profiles and results
After transcription choose the number of speakers (automatic, 2, 3 or 4) and Separate speakers. The app groups speech by voice features. Review the result, particularly similar voices, silence, short responses and overlapping speech.
- You can edit text, speaker names, speaker assignment and split utterances. Re-transcribing or separating again replaces the saved result; export a version you need first.
- A therapist profile helps label your own speech; it is not used to sign in or prove identity. Record 20–30 seconds in quiet conditions; creation becomes available after at least 10 seconds.
- A patient profile is optional and requires separate consent. It stores a name, date and numerical voice features. Delete it in the patient profile or Settings. Matching is not certain.
Privacy, backups and deletion
Recording and analysis do not send recordings, transcripts, patient records or voice profiles to the developer’s server or a cloud transcription service. Audio is analysed on the device. Data is in the app’s private storage: audio and sessions, transcripts, patient records, settings and voice profiles.
- The code excludes recordings and transcripts from system backup. It does not exclude patient records, settings or profiles stored in preferences; iOS device backup, including iCloud, may include them depending on your settings. The app has no own sync or cloud backup.
- Export is created only when you choose it. It may contain records, sessions, transcripts, profiles and audio; the app does not encrypt it. Sharing it through iCloud Drive, email or another service sends data to your chosen destination. Deletion in the app does not delete exports or backups.
- Deleting a session in its detail also deletes its transcript. Automatic deletion and Delete all recordings remove sessions and audio, but do not remove standalone transcript files, patient records or profiles. Manage these separately.
Failure, limits and licence
During recording the app writes segments and a local recovery checkpoint roughly every 10 seconds of active recording. After a failure it may recover readable segments as a session in History. This cannot guarantee all audio: an unclosed segment can be lost. Check duration and listen to the end; start further conversation as a new session.
- Transcription and speaker separation can be wrong. Always compare the result with the recording; the app does not replace professional judgement, consent duties or data-protection responsibilities.
- There is a 14-day trial and a one-time full-version purchase, without a subscription. After the trial, new recordings, transcription and speaker separation are blocked; history and export remain available. Price and Restore purchase are in Settings → Licence.
- If something fails, check microphone permission, free storage, recording language, licence and Internet for a first download. Settings → Data integrity runs diagnostics but does not repair or delete files. Contact: elsik@elsik.pl — do not send confidential recordings or transcripts.