---
license: mit
---
# Audio Feature Extraction Models

This repository contains pre-trained models for audio feature extraction, specifically:

- **Key Detection:** Classifies the musical key of an audio track into relative key classes.

## Model Details

### Tempo Model
- **Model Type:** Custom CNN architecture for tempo classification.
- **Input:** Audio segments converted to Mel spectrograms followed by autocorrelation.
- **Output:** Predicts Beats Per Minute (BPM) in a range from [85, 170].

### Key Detection Models
- **Key Class Model:** Classifies into 12 relative key classes.
- **Key Quality Model:** Determines if the key is Major or Minor.
- **Input:** Audio segments converted to Mel spectrograms.
- **Output:** 
  - Key Class: One of 12 key signatures.
  - Key Quality: Binary classification (0 for Major, 1 for Minor).

## Usage

### Prerequisites
- Python 3.7+
- PyTorch
- torchaudio
- transformers

### Loading Models

To use these models with Hugging Face's transformers library:

```python
from transformers import [AutoModelForAudioClassification](https://x.com/i/grok?text=AutoModelForAudioClassification)

# Load Tempo Model
tempo_model = AutoModelForAudioClassification.from_pretrained("your_username/tempo_model")

# Load Key Models
key_class_model = AutoModelForAudioClassification.from_pretrained("your_username/key_class_model")
key_quality_model = AutoModelForAudioClassification.from_pretrained("your_username/key_quality_model")