Amical setup: local voice input on Mac and Windows

This Japanese-language tutorial covers installing and using Amical for dictation on Mac and Windows. The speaker presents it as free to use and explains how it uses open source speech recognition technology. Suggested uses include drafting YouTube scripts, writing emails and dictating prompts during AI-assisted app development.

The setup starts on Amical's GitHub download page. The speaker identifies ARM64 for Apple Silicon Macs, a separate Intel Mac download, and Windows X64. Mac installers use DMG files; Windows uses an EXE. Inside the app, the tutorial covers model selection, language and microphone settings, launch at login, and a function-key push-to-talk shortcut.

Amical Cloud processes speech with a cloud model, while local options install and run models locally on the user's computer. The speaker prefers a large local model with a 3.1GB download and describes it as slower but more accurate according to the displayed ratings. Those ratings are not an independent benchmark, and the speaker's suggestion that local processing is safer is a personal assessment.

The dictation example produces usable text but misrecognizes DaVinci Resolve. The speaker then uses ChatGPT to tidy the transcript into a script. They report no noticeable quality loss compared with AquaVoice in their own use, while noting that speaking aloud can be awkward in public and that formal documents may need further editing.