First stable release of Mimora - a local, offline pronunciation trainer.
Highlights
- 100% offline after install: TTS, speech recognition, phrase generation and scoring all run locally. No cloud, no accounts.
- Word-level feedback: 0-100 score, the exact words to work on, reference-vs-you replay.
- Practice your own material: a local LLM turns any pasted text into an endless stream of phrases.
- Runs on a typical PC: 16 GB RAM, no GPU required (NVIDIA GPU optional).
Languages
- English (American, British) - calibrated phoneme engine (default), plus acoustic engine.
- Spanish (Castilian) - experimental, English calibration fallback until Spanish calibration lands.
Install
See Quick install - install.py handles dependencies, GPU builds and model downloads in one guided run (roughly 10 GB of disk once packages and models are in place).
Requires Python 3.11 or 3.12 and the native espeak-ng binary. Windows is the primary target; macOS (Apple Silicon and Intel) and Linux are supported.