Vibe Coding Discoverdasheng
View on GitHub大声读 — R2T2 流式 ASR 听，Jev 逐词判，英文朗读评分
A Next.js web app that scores English read-aloud: streaming R2T2 ASR transcribes speech, word-level alignment finds mismatches, and the Jev LLM answers closed questions (same word? error type? same sentence?) to flag errors and compute a weighted total. Self-hostable on a 12GB NVIDIA GPU or via a relay demo; pronunciat
Use Cases
English read-aloud scoringReal-time word-by-word ASR highlightingPer-word mispronunciation/misread detectionClosed-question LLM verdicts on transcriptsSelf-hosted offline speech assessment on a 12GB GPUMock ASR mode for testing without mic or GPU
Built With
- Language
- JavaScript
- Frameworks
- Next.js · React · undici
Tags
streaming-asr · speech-recognition · pronunciation-scoring · llm-judge · real-time · self-hosted · text-alignment · english-learning · websocket · pcm16 · voice · nextjs · audio