---
language: en
tags:
- audio
- noise-reduction
- speech-enhancement
- deep-learning
datasets:
- haydarkadioglu/speech-noise-dataset
license: mit
pipeline_tag: audio-to-audio
---


# Nocle - Noise Cleaner Model

## Model Description
This is a deep learning model trained for audio noise reduction, specifically targeting speech enhancement. The model is designed to effectively remove background noise while preserving the quality and intelligibility of speech.

## Intended Use
- Speech enhancement in noisy environments
- Audio cleanup for voice recordings
- Background noise reduction
- Audio quality improvement for voice applications

## Training Data
The model was trained using the [Speech Noise Dataset](https://huggingface.co/datasets/haydarkadioglu/speech-noise-dataset), which includes:
- Clean speech recordings
- Various environmental noise samples
- Mixed noisy speech samples for training

## Model Architecture
- Neural network optimized for audio processing
- Input: Noisy audio waveform
- Output: Enhanced clean audio waveform
- Processing: 16kHz sampling rate

## Usage
This model is integrated into the Nocle application. To use it:

1. Clone the repository:
```bash
git clone https://github.com/haydarkadioglu/nocle-app.git
cd nocle-app
```

2. Install dependencies:
```bash
pip install -r requirements.txt
```

3. Run the application:
```bash
python main.py
```

For programmatic usage:
```python
from setup import Setup
from model_handler import ModelHandler

# Model will be automatically downloaded
model = ModelHandler()
# Process audio
cleaned_audio = model.process_audio(noisy_audio)
```

For more detailed instructions and source code, visit the [GitHub Repository](https://github.com/haydarkadioglu/nocle-app).

## Performance and Limitations
### Strengths
- Effective at removing common background noise
- Preserves speech clarity
- Real-time processing capability

### Limitations
- Optimized for 16kHz audio
- Best suited for speech enhancement
- May require adjustment for non-speech audio

## Training Procedure
### Training Data
- Dataset: [Speech Noise Dataset](https://huggingface.co/datasets/haydarkadioglu/speech-noise-dataset)
- Audio format: 16kHz WAV files
- Mixed with various noise types at different SNR levels

### Training Parameters
- Optimizer: Adam
- Loss function: Mean Squared Error (MSE)
- Batch size: 12000 samples

## Ethical Considerations
- The model is designed for general noise reduction and should be used responsibly
- Users should respect privacy when processing audio containing personal information
- The model should not be used for deceptive audio manipulation

## Links
- [GitHub Repository](https://github.com/haydarkadioglu/nocle-app)
- [Training Dataset](https://huggingface.co/datasets/haydarkadioglu/speech-noise-dataset)
- [Model on Hugging Face](https://huggingface.co/haydarkadioglu/nocle-app)

## Citation
```bibtex
@misc{nocle2025,
  author = {Haydar Kadıoğlu},
  title = {Nocle: Noise Cleaner Model},
  year = {2025},
  publisher = {Hugging Face},
  howpublished = {\url{https://huggingface.co/haydarkadioglu/nocle-app}}
}
```