clip-tagger Model
This is a personalized audio tagging model based on CLAP (Contrastive Language-Audio Pre-training). It extends the base Xenova/clap-htsat-unfused model with user feedback and custom tags.
Model Description
- Base Model: Xenova/clap-htsat-unfused
- Framework: Transformers.js compatible
- Training: User feedback and custom tag integration
- Use Case: Personalized audio content tagging
Usage
import { CLAPProcessor } from './clapProcessor.js';
import { LocalClassifier } from './localClassifier.js';
// Load the model
const processor = new CLAPProcessor();
const classifier = new LocalClassifier();
classifier.loadModel(); // Loads from localStorage or model files
// Process audio
const tags = await processor.processAudio(audioBuffer);
const personalizedTags = classifier.predictAll(features, candidateTags);
Files
localClassifier.js
- Local classifier implementationclapProcessor.js
- CLAP model wrapperuserFeedbackStore.js
- User feedback storage systemmodel-config.json
- Model configurationexample-usage.html
- Usage example
Links
- π Live Demo: clip-tagger Space
- π¦ Model Repository: sohei1l/clip-tagger
- π» Source Code: GitHub Repository
Training Data
This model learns from user corrections and custom tags. The base CLAP model provides initial audio understanding, while the local classifier adapts to user preferences.
Inference Providers
NEW
This model isn't deployed by any Inference Provider.
π
Ask for provider support