SlowTalk: a patient language tutor that never leaves my friend's laptop

Description and use

SlowTalk is a local, privacy‑first language tutor that runs open models via Ollama and Whisper, enabling safe speaking practice without sending data online.

Problem

A language learner struggles with speaking and fears that online tutors send their voice or text to external servers, compromising privacy and causing anxiety.

Outcome

SlowTalk provides a private, patient language tutor that runs entirely offline, keeping all user data on the laptop and allowing the learner to practice speaking without fear of data leakage.

Workflow

• Install Python 3.11+ and add to PATH
• Install Ollama and pull the gemma3:4b model via ollama pull gemma3:4b
• Edit friend.json with name, language, level, interests, and optional model
• Double‑click run.bat to start the Flask server and open the web interface
• Use the Speak button to record a sentence, which Whisper transcribes locally
• The tutor replies in the target language, short sentences, ending with a question
• Corrections appear as a pink TIP line with a brief English explanation
• Read aloud feature uses the browser’s speech synthesis at 0.85 speed

Evidence

Everything about [FRIEND] lives in one small file, friend.json: name, language, level, hobbies and goal.

Source

It runs an open-weight model locally through Ollama, so there is no account, no API key, no cost, and no internet needed after setup.

Source

The brain: Gemma 3 (4B) through Ollama.

Source

Voice in: Whisper, running locally.

Source

The first start downloads the models, so it needs internet once.

Source

What surprised me: she turned on Read aloud within a minute and kept repeating the tutor's sentences back to herself.

Source