fxy2311-youyou/expression-trainer

What it solves

This project provides a tool for improving oral expression precision. It helps users identify and reduce the use of filler words, hesitant phrases, and vague terminology during speech, providing real-time feedback and detailed analysis to help users speak more clearly and directly.

How it works

The application uses a local Electron-based desktop environment. It integrates Sherpa-ONNX for completely offline, real-time speech-to-text recognition. The recognized text is then matched against a local lexicon (based on the Dalian University of Technology emotional lexicon) to flag filler words and vague expressions. For deeper analysis, the system connects to AI backends (such as DeepSeek, OpenAI, or Ollama) via HTTP APIs to generate comprehensive feedback reports based on six dimensions: logic, directness, filler words, density, vocabulary, and highlights.

Who it’s for

People looking to improve their public speaking, oral communication skills, and the precision of their spoken language, particularly those who prefer local processing for privacy or offline capabilities.

Highlights

• Offline ASR: Uses Sherpa-ONNX for real-time, offline speech recognition.
• Real-time Visual Feedback: Displays subtitles with color-coded underlines (red for fillers, orange for hesitant words, yellow for vague words, green for powerful expressions).
• Multi-Backend AI Support: Compatible with Groq, OpenAI, DeepSeek, and Ollama.
• Detailed Analysis: Provides a 6-dimensional deep analysis report upon completion of a session.