This guide covers every section of the Settings modal. Open Settings at any time using the floating gear icon, which is available on every screen. Click « Save Settings » to apply changes, or « Cancel » to close without saving.
A. Transcription settings
These settings control how the transcription engine chunks and processes audio.
- Chunk Length (seconds) — how large each transcription chunk is. Larger chunks can improve continuity but may increase memory pressure.
- Stride Length (seconds) — overlap between chunks. Stride helps avoid missing words at chunk boundaries. If you use word-level timestamps, overlap is disabled for that job regardless of this setting.
- Boundary Merge Sensitivity — controls how aggressively the app merges transcript text at overlapping chunk boundaries. Options are Very Aggressive, Aggressive, Moderate, Conservative, and Prudent. Start with Moderate; adjust if you see repeated or missing words at chunk joins.
- Merge segments by sentence — automatically merges text segments by sentence after the transcription job ends.
- Autosave transcription edits — when enabled, the app saves transcript changes automatically after edits. When disabled, you must click the Save icon manually.
- Default Language — pre-fills the language field in the upload and edit modals so you do not have to set it each time.
B. Waveform settings
- Waveform below transcription — shows or hides the waveform panel below the transcript editor. Useful for navigation but adds visual workload and some extra processing.
- Waveform scroll speed — controls how fast the waveform scrolls during playback.
Tip: If transcription or editing feels slow, try disabling the waveform. It is not required for any core workflow.
C. Display settings
- Theme — switches between available visual themes.
- Single column transcription view — toggles a single-column editor layout. The same toggle also exists in the transcript toolbar.
- Headings & Labels Font — changes the font used for headings and labels.
- Main Text Font — changes the font used for transcript body text.
D. Compute settings
These settings control which hardware and precision mode the app uses when loading and running a model.
- Compute Device — choose CPU or GPU. GPU is often faster but only if the machine and dtype combination support it. CPU is slower but more predictable.
- GPU Backend — selects the GPU backend (hidden on macOS Apple Silicon, which uses a CoreML-first route).
- GPU Power — available on supported systems only (hidden on macOS).
- CPU Cores — the number of CPU threads used for CPU-based inference.
- Graph Optimizations — lower levels can help if model loading is unstable. Try reducing this setting if you experience repeated loading failures.
- GPU Default Dtype — the default precision or quantization mode applied when using the GPU. « auto » is the safest starting point.
- CPU Default Dtype — the same choice for CPU-based inference.
Dtype quick reference:
- auto — best starting point; lets the app choose the right mode for your device.
- fp32 — highest precision, highest memory cost, often slower.
- fp16 — lower memory and often faster on supported GPUs.
- uint8 / int8 / q8 — quantized options with smaller downloads and lower memory use.
- q4 / q4f16 / bnb4 — more aggressive quantization to reduce size and memory further.
- none / raw — minimal compression, largest footprint.
E. Storage settings
- Transcriptions Folder — the folder where the app stores all project data. Click « Browse » to choose a custom location.
- Custom Models Folder — the folder where the app looks for locally stored model weights. Click « Browse » to set it (subscription-gated), then click « Scan » to detect the models in that folder.
Important: Do not store your active Transcriptions Folder inside iCloud Drive, Dropbox, OneDrive, Google Drive desktop sync folders, or network-mounted cloud folders. Live sync tools can delay, lock, duplicate, or overwrite the JSON files the app writes while you work. Keep active work on a normal local folder and use Export Project for backups.
F. Subscription settings
This section handles email registration and device linking for your subscription.
- Register Email — enter your email and click this to begin activation. If a verification code is required, a code input appears.
- Paste — pastes a copied OTP code into the code field (visible when waiting for a code).
- Verify — submits the 6-digit code to complete activation (visible when waiting for a code).
- Buy Subscription — opens the purchase flow (visible when no active subscription is found).
- Manage Subscription — opens the subscription management URL for billing changes (visible only when a manage URL is available).
- Unlink Device — detaches this machine from your subscription without affecting other devices.
- Reset all devices — opens a modal that sends a reset link to your registered email if you have hit the device limit. The link expires in 2 hours.
- Clear — removes subscription state from the app on this device only.
G. Cached Models
This section shows the model weights that have been downloaded and stored locally. Managing this list does not affect your transcripts or notes.
- Refresh — rescans local storage and updates the cached model list.
- Per-model delete control — removes a specific cached model variant from the list.
- Clear All Cached Models — deletes all locally cached model files. Models will be re-downloaded the next time you transcribe.