Settings

Model

Model Path

Directory where Whisper model files are stored. Auto-downloaded to default path on first use; you can also specify manually.

Model Selection

Whisper Models

• tiny: Fastest, lowest accuracy, good for testing
• base: Fast, moderate accuracy, good for daily use
• small: Balanced choice, recommended for beginners
• medium: High accuracy, moderate speed
• large: Highest accuracy, slowest, needs more RAM
• turbo: Optimized large, faster speed

Qwen3-ASR Models

• 0.6B: ~1.7 GB, lightweight and fast, recommended for Chinese
• 1.7B: ~4.5 GB, higher accuracy on complex audio, Pro only

You can switch the ASR engine in Settings → Model. Qwen3-ASR is better for Chinese, dialects, and Chinese-English code-switching. See Qwen3-ASR Engine.

GPU Acceleration

• macOS: Auto uses CoreML / ANE (Apple Silicon)
• Windows/Linux: Select GPU device index, Vulkan supported

Transcription Options

Language

• auto: Auto-detect (slightly lower accuracy)
• Specified: e.g. zh (Chinese), en (English), ja (Japanese)

Temperature

Controls sampling randomness. 0 = fully deterministic, 1 = most random. Default 0, recommended to keep.

Stable Timestamps

Silero VAD-backed precise timestamp mode. More accurate subtitle timing when enabled, slightly slower processing.

Speaker Diarization

Requires additional ONNX model download when enabled. Supports 2-4 speaker separation.

Recording

• Default Device: Set default microphone and system audio device
• Recording Format: WAV (lossless) or MP3 (compressed)
• Save Path: Default save location for recordings
• Auto-transcribe: Whether to auto-start transcription after recording

AI

LLM Provider

Choose your AI service:

• Ollama: Local, data never leaves device
• Claude / DeepSeek / Gemini / Qwen / Zhipu: Cloud API, requires Key

API Key

Get your API Key from the provider’s website and enter it here. Key is saved locally only, never uploaded.

Custom Prompt

Edit default prompt templates for summary, chat, polish, etc.

Export

• Default Format: Set default export format
• Include Timestamps: Whether to include time info in exports
• Encoding: UTF-8 or UTF-8-BOM

Shortcuts