cyx2333hhh/talk-type

What it solves

Talk-type provides a seamless voice-to-text experience for macOS users, specifically addressing the challenges of bilingual (Chinese and English) input and the need for polished, grammatically correct text that can be inserted directly into any active application.

How it works

The tool uses a multi-layered recognition pipeline. It first utilizes Apple Speech for real-time transcription. If that fails, it falls back to a local Whisper Small model via whisper-cli. For bilingual support, it performs a second pass with the alternative language to ensure accuracy. Once transcribed, the text can be processed by an optional AI cleanup layer (supporting providers like OpenAI, Claude, and DeepSeek) that uses surrounding cursor context to fix punctuation and sentence breaks before inserting the text into the focused app using macOS Accessibility APIs.

Who it’s for

It is designed for macOS users who frequently dictate text for writing, messaging, or note-taking and need a tool that handles mixed-language input and AI-driven text refinement.

Highlights

• Bilingual Support: Native handling of Chinese, English, and mixed-language input.
• Local Fallback: Integrates whisper.cpp for offline/local transcription when system speech recognition fails.
• AI-Powered Cleanup: Supports multiple LLM providers to refine transcripts into polished text.
• Cross-App Integration: Directly writes text into any focused input field using macOS Accessibility features.