Skip to content

PClmnt/Mimicly

Repository files navigation

Mimicly

Mimicly is an AI-assisted pronunciation practice app for language learners who want tight feedback loops instead of passive flashcards.

It generates compact speaking lessons around a scenario, lets learners hear each phrase, record an attempt, and get immediate scoring and coaching. Missed phrases are saved into a review queue so weak spots stay visible until they improve.

Features

Focused lesson generation

  • Builds short, scenario-based speaking lessons for a chosen language, topic, and difficulty.
  • Keeps lessons compact so practice stays deliberate instead of sprawling.
  • Adds short phrase-level cues to help learners focus on rhythm, stress, or pronunciation.

Pronunciation-first practice loop

  • Plays phrase audio before each attempt so learners hear the target clearly.
  • Records the learner's response and scores the attempt against the target phrase.
  • Returns direct feedback, a score, a short next step, and the transcription the system heard.

Review memory

  • Automatically saves weak phrases into a review queue.
  • Keeps review focused on phrases that actually caused trouble.
  • Removes phrases from the queue only after strong repeated performance.

Faster, lower-friction experience

  • Shows the lesson immediately instead of blocking on non-essential media generation.
  • Prefetches phrase audio in the background for the current and next card.
  • Treats scene images as optional memory aids rather than required content.

Clean session flow

  • Replaces the previous split demo modes with one clear product path.
  • Supports fresh practice sessions, resuming the latest lesson, and targeted review sessions.
  • Preserves recent lessons and local practice progress between visits on the same device.

Multi-language support

  • Supports Chinese, Japanese, Spanish, French, Korean, German, Italian, and Portuguese.
  • Uses language-aware transcription settings so scoring is aligned with the selected language.
  • Includes romanization only when it is useful for the target language.

Tech Stack

  • React 19 for the client application
  • TypeScript across frontend and Worker code
  • Bun for package management and script execution
  • Vite for local development and production builds
  • Hono for the Worker API layer
  • Cloudflare Workers for deployment and edge execution
  • Cloudflare D1 for lesson history, attempts, and review state
  • Cloudflare Workers KV for lesson template caching
  • Cloudflare R2 for generated audio and scene image storage
  • OpenAI APIs for text-to-speech and optional scene image generation
  • Mistral APIs for transcription, lesson generation, and pronunciation evaluation

Product Summary

Mimicly is designed around one idea: speaking practice should be short, immediate, and repeatable.

Instead of generating content and leaving the learner on their own, it pushes each session toward a simple loop:

  1. Generate a useful lesson.
  2. Hear the phrase.
  3. Say it back.
  4. Get scored.
  5. Revisit what was weak.

About

Voice to Text language learning application using Cloudflare workers / Mistral Voxtral

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors

Languages