—
How It Works
AI Model
Uses onnx-community/Speech-Emotion-Classification-ONNX - a wav2vec2 model fine-tuned on RAVDESS for 8-class emotion recognition, exported to ONNX for in-browser inference.
Emotion Labels
The model classifies speech into: neutral, calm, happy, sad, angry, fearful, disgust, and surprised. All eight labels are shown with confidence scores.
On-Device Inference
The ONNX model runs entirely in your browser via @xenova/transformers. Your audio never leaves your device.
Limitations
Emotion detection from voice alone is imperfect. Results reflect acoustic patterns - tone, pace, energy - not necessarily true emotional state.