jayjiratta/Speech-Text-Emotion-Analysis — reverse-engineered prompt

Build me a simple local web app called VoxSent that lets someone understand the emotion in either typed text or an uploaded audio file.

The user should be able to type a message, or upload speech audio. If it’s audio, transcribe it with Whisper first, then analyze the text. Let the user choose between a detailed 28 emotion model and a simpler 7 emotion model with anger, disgust, neutral, surprise, joy, sadness, and fear. Show the predicted emotion clearly, include a confidence score if available, and make the result feel friendly with emojis or a simple visual summary.

Please also add an about or insights page that explains the two models in normal language and shows training results like confusion matrices, reports, and model charts when those files are available. Use the existing trained models and labels in the project. Make the Streamlit interface clean, easy to run locally, and include any setup notes needed for audio transcription.