mimora

Mimora is an offline, privacy-focused pronunciation trainer designed for language learners and accent reduction. It operates entirely on the user's machine, utilizing local models for text-to-speech (TTS), speech recognition, and phrase generation. The core workflow involves a local LLM generating phrases from user-provided text, which are then spoken by a consistent TTS voice. The user repeats the phrase, and Mimora analyzes the recording against the reference audio, providing a score out of 100 and highlighting specific words that need improvement. The app supports English (American and British) and Spanish (Castilian), with English being fully calibrated and Spanish currently experimental. Key features include one-press recording, automatic silence detection, and the ability to replay reference and user recordings for comparison. Mimora offers two scoring engines: a default 'phoneme' engine that compares IPA phonemes, and an 'acoustic' engine that compares audio embeddings. It runs on standard hardware (16 GB RAM, no GPU required) and supports Windows, macOS, and Linux. Installation is recommended via `uv` or `pipx`, with models downloaded on first run. The app includes a settings window for adjusting language, accent, voice, speed, and translation options, and supports a 'No-LLM mode' for slower machines where phrases are taken verbatim from the user's text.