Skip to content

README.md

Latest commit

 

History

History
81 lines (56 loc) · 3.04 KB

README.md

File metadata and controls

81 lines (56 loc) · 3.04 KB
Raw

Emotion Analyzer

Visualize speaker emotions across a conversation. Upload any recording and see who said what — and how they felt saying it — powered by Smallest AI Pulse STT emotion detection.

Demo

Demo

Features

  • Upload audio and get a diarized transcript with per-segment emotion scores
  • Interactive timeline chart with one line per speaker–emotion combination
  • Color-coded emotion filters (happiness, sadness, anger, fear, disgust)
  • Speaker filters with dash-style line indicators
  • Concurrent emotion detection for fast results on long recordings

Requirements

Base dependencies are installed via the root requirements.txt. See the main README for setup. Add SMALLEST_API_KEY to your .env.

Extra dependencies:

uv pip install -r requirements.txt

Usage

uv run backend/app.py

Open http://localhost:5000 in your browser.

  1. Drop or select an audio file
  2. Click Analyze Emotions
  3. Explore the results — emotion timeline, speaker filters, and full transcript table

Recommended Usage

  • Analyzing tone and sentiment shifts across a multi-speaker conversation
  • Comparing emotional patterns between speakers over time
  • For basic emotion scores without visualization, File Transcription supports emotion_detection=true directly

Structure

emotion-analyzer/
├── .env.sample          # Environment variable template
├── requirements.txt     # Extra Python dependencies (flask, pydub)
├── README.md
├── backend/
│   └── app.py           # Flask backend — upload, diarize, merge, split, emotion detect
└── frontend/
    ├── index.html       # Minimal HTML shell
    ├── app.js           # UI logic
    └── style.css        # Styles and theming

How It Works

  1. Upload — the Flask backend receives an audio file via /api/analyze
  2. Diarize — sends the full audio to Pulse STT with diarize=true and word_timestamps=true, getting speaker-labeled utterances
  3. Merge — utterances sharing the same timestamp and speaker are merged into single segments
  4. Split — pydub slices the original audio at utterance boundaries into per-segment WAV clips
  5. Detect emotions — each segment is sent to Pulse STT with emotion_detection=true concurrently (up to 10 workers)
  6. Visualize — the frontend renders an interactive Chart.js timeline with emotion/speaker filter chips and a transcript table

Supported Formats

Audio: WAV, MP3, FLAC, OGG, M4A, AAC, WMA

API Reference

Next Steps