Overview

BSRoformer GGUF is a specialized audio-to-audio model optimized for speech enhancement and noise reduction. By leveraging the GGUF format, this model is designed for efficient local deployment, allowing developers to integrate high-fidelity audio cleaning directly into edge applications without requiring massive GPU clusters. It excels at isolating target speech from complex background noise, making it an ideal component for real-time transcription pipelines, VoIP software, or accessibility tools. Compared to standard PyTorch weights, the GGUF quantization significantly lowers the memory overhead and improves inference speed on CPU-bound environments, providing a pragmatic balance between audio quality and computational cost.

Highlights

Optimized GGUF format for efficient local CPU/GPU inference
High-performance speech enhancement and background noise removal
Low memory footprint for edge device integration
Apache-2.0 license for flexible commercial deployment
Ideal for pre-processing audio before ASR pipelines

Usage

Install

# Install Hugging Face transformers
pip install transformers torch

SDK Usage

# Load model with transformers
from transformers import AutoModel, AutoTokenizer

model = AutoModel.from_pretrained("chenmozhijin/BSRoformer-GGUF")
tokenizer = AutoTokenizer.from_pretrained("chenmozhijin/BSRoformer-GGUF")

Hugging Face Download

We recommend downloading the model via the Hugging Face CLI or Hub SDK.

Guidance：Before downloading, install huggingface_hub with:

Guidance

pip install -U huggingface_hub

CLI Download

Download the full repository

huggingface-cli download chenmozhijin/BSRoformer-GGUF

Download a single file to a local folder (e.g. config.json into ./dir)

huggingface-cli download chenmozhijin/BSRoformer-GGUF config.json --local-dir ./dir

See the official docs for more CLI options

SDK Download

# 模型下载
from huggingface_hub import snapshot_download
model_dir = snapshot_download('chenmozhijin/BSRoformer-GGUF')

Git Download

Make sure git-lfs is installed first

Git Download

git lfs install
git clone https://huggingface.co/chenmozhijin/BSRoformer-GGUF

To skip LFS large-file downloads, use:

Skip LFS

GIT_LFS_SKIP_SMUDGE=1 git clone https://huggingface.co/chenmozhijin/BSRoformer-GGUF

Model files are hosted on the Hugging Face Hub — download directly via HF CLI / SDK / Git, not through this site.

PyTorch / Transformers Usage

Install Transformers

pip install -U transformers torch

Load the model and run inference

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained('chenmozhijin/BSRoformer-GGUF')
tokenizer = AutoTokenizer.from_pretrained('chenmozhijin/BSRoformer-GGUF')

Full Documentation

来源: HuggingFace

---
tags:

audio

music

source-separation

mel-band-roformer

pipeline_tag: audio-to-audio
---

BSRoformer-GGUF

Official GGUF model repository for the BSRoformer.cpp project.
This repository contains BS Roformer/Mel-Band-Roformer models converted to the GGUF format, enabling high-performance, cross-platform local inference using the MelBandRoformer.cpp inference engine.

📦 Model List

Quantization Types

To meet different hardware requirements, various quantized versions are provided:

q8_0 (Recommended): 8-bit quantization. Significantly reduces memory usage and bandwidth requirements while maintaining audio quality almost identical to FP32.

fp16: 16-bit floating point. Suitable for scenarios requiring maximum precision.

q4_0 / q4_1 / q5_0 / q5_1: Lower-bit quantization, suitable for devices with low VRAM/RAM.

*(Note: Norm and Bias weights remain in FP32 to ensure numerical stability)*

🚀 Usage

Please use with the BSRoformer.cpp command-line tool.

1. Download Tool

Download the executable for your system from BSRoformer.cpp (or compile it yourself).

2. Download Model

Download the desired .gguf file (e.g., voc_fv6-Q8_0.gguf) using the direct links in the Model List above, or browse all models from the Files and versions page of this repository.

3. Run Inference

```bash

Basic usage

./bs_roformer-cli model_q8_0.gguf input.wav output.wav

Advanced usage (adjust chunk size and overlap to optimize quality)

./bs_roformer-cli model_q8_0.gguf input.wav output.wav --overlap 2