Skip to main content

​
Settings

Open settings from the gear icon in the main window. Settings are organized into the following sections.

​
Account

Signing in is entirely optional and only needed for Pro features - the Free tier transcribes fully offline with no account. If you do sign in, view your email, linked providers (Google, email), and current plan tier. You can enter a license key, link additional sign-in providers, or sign out.

​
Audio Input

  • Input Device: pick your microphone, or leave it on System Default
  • Recording Mode: Push to Talk (default) or Open Mic - see Input Modes
  • Open Mic Threshold (0-100%): only shown in Open Mic mode. The minimum voice level for open mic to start recording. Use Test to watch a live audio meter. Turns red and warns you past 80%, since a high threshold can miss quiet or soft-spoken starts.

​
Recording

  • Max duration: 5 seconds–5 minutes per recording (default 30 seconds; same ceiling on every tier, no account required to change it). The slider fades from green to yellow to red past 1 minute - transcription isn’t streaming, so longer recordings take longer to process and show nothing until they finish.
  • Remove filler words: Filter out “umm”, “uh”, “like”, etc. Pro
  • Normalize numbers to digits: Convert “three” → “3” Pro
  • Append space to incomplete sentences: Adds a trailing space when a transcription doesn’t end with ., !, or ?, so you can pause mid-sentence and keep dictating without manually adding a space (on by default)

​
Hotkeys

Set the global Toggle Recording hotkey. It works globally whenever the app is running. Default: Ctrl+Shift+Space on Windows/Linux, or the Dictation key (mic icon, shown as F5) on macOS. Click the binding to enter capture mode, then press your desired key combination. ESC cancels.
ESC and Spacebar also work as recording toggles when the app window is focused.

​
Voice Model

View and change the active transcription model. Shows your detected GPU and the active model. Click Change to browse and download other models.
Available models depend on your tier. See the beta page for a comparison.

​
Custom Model Pro

Bring your own ONNX transcription model - import a local .onnx file or pull one straight from a HuggingFace repo URL. Echophrase validates the model before switching to it. See Custom Models for supported formats and where to find models.

​
Transform Model Pro

Enable an optional post-processing step that applies grammar and punctuation correction to transcriptions. Uses the lightweight T5-Mini Grammar model (~60 MB), downloaded on demand and run entirely on-device. This is a Pro-only feature - the toggle is locked on the Free tier. You can write custom transformation rules or use the default grammar correction.

​
GPU Acceleration

Echophrase auto-detects the fastest available execution provider at runtime. The Auto (Recommended) option picks the best device for your hardware: DirectML on any Windows DirectX 12 GPU, CoreML on Apple Silicon, or NPU on Copilot+ PCs. CPU is the fallback when nothing else is available. You can force a specific provider if you need to:
  • Auto (Recommended): Let Echophrase pick the fastest device at session start
  • DirectML: Force Windows DirectML (any DirectX 12 GPU)
  • CoreML: Force Apple Metal (macOS 13.3+ on Apple Silicon)
  • CUDA (Windows, NVIDIA, opt-in): Download the CUDA plugin and force the CUDA provider - requires CUDA Toolkit 13.0+ and cuDNN 9.19+, and a restart
  • CPU: Force CPU - useful for battery savings or when the GPU is busy elsewhere
DirectML covers NVIDIA GPUs on Windows with no driver/toolkit version juggling, so it is the default. NVIDIA users with CUDA 13.0+ and cuDNN 9.19+ installed can opt in to a downloadable CUDA plugin here. See GPU Acceleration.
Changing the active provider requires restarting Echophrase to take effect. The settings UI shows a “Restart to apply” banner when a change needs a relaunch.

​
Output

Control what happens after a transcription completes.
  • Format: Plain Text (all tiers), JSON with custom metadata template Pro
  • Post-record actions (choose one or more):
    • Copy to Clipboard (default)
    • Paste at Cursor (free for everyone, but requires signing in - it’s the one Free-tier output action gated behind an account rather than Pro; everything else on the Free tier, including transcription itself, stays fully account-free)
    • Save as File Pro - configure an auto-save directory and export transcripts, including JSON with custom metadata Pro

​
Language

Choose the UI language. Currently supported: English and Traditional Chinese (beta).

​
Customize

Toggle visibility of UI elements Pro:
  • Show Hotkey Changer
  • Show Transcription Preview
  • Show Output Confirmations
  • Show Copy Last Button
  • Show Input Changer
  • Show Model Switcher
  • Show Transcription Speed in History
  • Hover to Reveal Last Transcription
  • Minimize to Tray on Close

​
Accessibility

Scale the entire interface to your preference - useful on high-DPI displays or if the default size is hard to read.
  • UI Scale - Auto (default): Matches your operating system’s display scaling (e.g. 150% Windows scaling)
  • UI Scale - Manual: Set any scale from 50% to 200% with a slider; changes apply live
  • Scale UI with window size: Optionally grow or shrink the interface as you resize the window
  • Reset scale and window: Restore the default scale and window size
In rare environments where OS scaling can’t be detected reliably, Auto falls back to 100% and the app suggests switching to Manual.

​
Cloud Sync Pro

Opt in, per device, to syncing your application settings to our servers so they follow you across every machine you sign in on. Off by default. Settings only - voice data, audio, and transcriptions never sync, and neither do API keys, hardware settings, or local file paths. You can push this device’s settings up or pull the server’s copy down at any time.

​
MCP Server Pro

Enable this toggle to let AI agents (like Claude Code) drive Echophrase’s recording and transcription through the CLI’s MCP server mode. Off by default and requires an active Pro subscription. The desktop app must stay running for the whole agent session, since it owns the loaded transcription model. See CLI and MCP Server for the full tool list and setup.

​
Data

  • Send anonymous app stats: toggle anonymous crash reports and usage stats (no audio or text is ever collected). Asked on first run, pre-checked; every event honors it, and DO_NOT_TRACK=1 overrides it. See What We Collect
  • Get Data: GDPR data export (rate-limited to once every 30 days)
  • Delete Account: Anonymizes your account data and signs you out (processed within a separate 30-day SLA - a different 30-day window than the export rate limit above)
  • Legal: Links to Terms of Service and Privacy Policy

​
About

View the current app version, check for updates, and access quick links to the website, support, docs, and Discord.