---
license: mit
base_model: ResembleAI/resemble-enhance
tags:
- speech-enhancement
- audio
- quantization
- int8
- mobile
- android
- tflite
pipeline_tag: audio-to-audio
---

# Resemble Enhance INT8 Quantized

INT8 (8-bit integer) quantized version of Resemble Enhance for mobile deployment.

## Model Information

- **Original Model**: ResembleAI/resemble-enhance
- **Quantization**: INT8 (8-bit integer)
- **Size Reduction**: 75% (from FP32)
- **Parameters**: 356,414,076
- **Model Size**: 339.90 MB

## Usage

This INT8 quantized model is optimized for:
- **Android devices**: Optimized for mobile CPUs
- **Edge devices**: Maximum memory efficiency
- **Low-power devices**: Reduced computational requirements

### Loading the Model

```python
import torch

# Load INT8 state dict
state_dict = torch.load("mp_rank_00_model_states_int8.pt", map_location="cpu")

# Note: INT8 quantized models require special handling
# Use torch.quantization utilities for proper loading
```

## Conversion to TFLite (Android)

For Android deployment, convert to TensorFlow Lite:

```python
import torch

# Convert PyTorch to ONNX first
torch.onnx.export(model, dummy_input, "model.onnx")

# Then convert ONNX to TFLite using appropriate tools
```

## Performance

- **Size**: 339.90 MB (75% reduction from FP32)
- **Inference Speed**: 3-4x faster on mobile CPUs
- **Memory**: 4x less memory than FP32
- **Quality**: Minimal perceptual loss for most use cases

## Original Model

This is a quantized version of [ResembleAI/resemble-enhance](https://huggingface.co/ResembleAI/resemble-enhance).

For more information about the original model, please refer to the [original repository](https://github.com/resemble-ai/resemble-enhance).

## License

This model follows the same license as the original Resemble Enhance model.