Turn ChatGPT into a Desktop Voice Tutor

What it does

Polyglot is an Electron-based desktop app—also available in a web build—that lets you practice spoken language with AI personas. It leans on OpenAI models for conversation and Azure services for text-to-speech and voice recognition. You supply your own API keys, choose a language and character, and start talking or typing.

The interesting bit

The app treats conversation like a rehearsal: Azure TTS highlights each word as it is spoken, you can replay your own voice to audit your pronunciation, and the spacebar serves as a push-to-talk shortcut. Custom dialogue scenes and AI-suggested replies keep the drill from turning into awkward silence.

Key highlights

• Supports English out of the box; other languages and AI personas are user-defined.
• Word-level highlighting synced to Azure TTS audio playback.
• Push-to-talk voice recognition (spacebar) with playback of your own recordings.
• Includes text translation, dark mode, and reply suggestions from the AI.
• Cross-platform desktop builds for macOS, Windows, and Linux; web version exists but mobile compatibility is noted as poor.

Caveats

• Requires your own OpenAI and Azure API keys; no managed backend is provided.
• macOS Gatekeeper blocks the app by default because Apple cannot verify it; the FAQ documents a manual workaround.
• The web version’s mobile compatibility is explicitly described as low.

Verdict

Worth a look if you want a configurable, voice-first AI tutor and are comfortable managing cloud API keys. Look elsewhere if you need a polished native mobile app or a fully offline stack.