Important: The maintainers(s) of this repository are not affiliated or connected with the original version of VoiceFixer.

Note: We are actively accepting contributions (+ contributors)! Please check the To Do list for how you can contribute!

VoiceFixer 2

Welcome to VoiceFixer 2, the next generation of VoiceFixer. VoiceFixer is a general speech restoration tool, using AI to remove background noise, fix degraded speech, enhance audio quality from old recordings, upscale audio resolution, and more, all in one model!

VoiceFixer aims to restore human speech, regardless of how seriously degraded it is. It can handle noise, reverberation, low resolution, and clipping effect within one model!

What's different from the original VoiceFixer?

The original version of VoiceFixer continues to be updated with minor changes and bug fixes, however if one tries to install it and run it out of the box, one would encounter several errors that require modifying installed packages to fix.

What’s the problem? How does this fix it? VoiceFixer requires an old version of the librosa library, which is incompatible with new versions of the numpy library. We’ve fixed this issue by fixing the old version of librosa and voicefixer. We also added several new features.

New features in VoiceFixer 2

We’ve added the following features in VoiceFixer 2:

We’ve added MPS support, which means you can use GPU acceleration on M1 macs. You can enable this by setting the cuda parameter to True. It’s automatically enabled when using the command line interface (CLI).
We've added a progress bar through TQDM for longer audio
We now support non-WAV files (ie MP3)
We're now using cached_path instead of hard-coding a cache path to increase OS support
We're featuring faster model downloads w/ Hugging Face
More features coming soon!

Changelog

Nov 18, 2023: Fix issue with model cache, thank to @gkarmas. Issue caused by spelling error 😳
Nov 16, 2023: Upgrade librosa + torch
Nov 11, 2023: Publish to PyPI
Nov 11, 2023: Add progress bar support (requires ffmpeg) (see TODO below)
Nov 11, 2023: Add preliminary MP3 support (requires ffmpeg) (see TODO below)
Nov 11, 2023: Fix CLI issue (see TODO below)
Sep 14, 2023: Switch to NOSCL-C-2.0 license
Sep 11, 2023: Forked from VoiceFixer

To-Do

Here's what we still need to do - feel free to contribute:

Demo

Check out the demos to see what VoiceFixer can do!

Installation

Don't want to install the package, but just want to try it out?

Use our free API (no API key required) for audio files under 5 minutes. Non-commercial use only, audio may be collected. Details on webpage.

curl -X POST -H "Content-Type: multipart/form-data" -F "file=@test.mp3" https://voicefixer-voicefixer-api.hf.space/process_audio > processed_audio.wav

NOTE: If you have any issues on Apple Silicon, please install PyTorch Nightly (pip install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cpu)

You can install our package via. PyPI (Python Package Index), the official Python package index.

pip install voicefixer2

This will install the latest published release.

If you would like to install the latest development version, or do not trust PyPI for any reason, please install directly from the source:

pip install git+https://github.com/fakerybakery/voicefixer

Including in Requirements

You may include voicefixer2 in your requirements.txt file:

voicefixer2

git+https://github.com/voicefixer/voicefixer

or, in setup.py

[
    'voicefixer2 @ git+https://github.com/voicefixer/voicefixer',
]

or simply

[
    'voicefixer2',
]

FFmpeg

NOTE: For MP3/OGG/etc (non-WAV) support, you must install FFmpeg

Quick Installation

macOS: brew install ffmpeg
Linux/Ubuntu: sudo apt install ffmpeg
Windows: scoop install main/ffmpeg

This is not guaranteed to work on all devices. Please see FFmpeg's website for instructions to install manually.

Usage

Important: FFmpeg must be installed to support non-.wav files.

Command Line

By default, if no output path is specified, the file will be saved to outfile.wav.

Process a file:

voicefixer --infile test/utterance/original/original.wav

Process all files in a directory:

voicefixer --infolder /path/to/input --outfolder /path/to/output

Change modes (default 0):

voicefixer --infile /path/to/input.wav --outfile /path/to/output.wav --mode 1

Run all modes:

voicefixer --infile /path/to/input.wav --outfile /path/to/output.wav --mode all

For more information:

voicefixer -h

Python API

from voicefixer import VoiceFixer
voicefixer = VoiceFixer()
# or voicefixer = VoiceFixer(model='voicefixer/voicefixer')
# Mode 0: Original Model (suggested by default)
# Mode 1: Add preprocessing module (remove higher frequency)
# Mode 2: Train mode (might work sometimes on seriously degraded real speech)
for mode in [0,1,2]:
    print("Testing mode",mode)
    voicefixer.restore(
        input=os.path.join(git_root,"test/utterance/original/original.flac"), # low quality .wav/.flac file
        output=os.path.join(git_root,"test/utterance/output/output_mode_"+str(mode)+".flac"), # save file path
        cuda=False, # GPU acceleration
        mode=mode
    )
    if (mode != 2):
        check("output_mode_" + str(mode) + ".flac")
    print("Pass")

License

VoiceFixer 2 is licensed under the VoiceFixer license, a license based on the BSD-3-Clause license with an additional restriction on the logo. Although it is not approved by the OSI, it follows the OSI's guidelines for an open source license.

Contributions to this software, including but not limited to pull requests, issues, suggestions, or code contributions, are subject to the following terms:

By submitting contributions to this project, you grant the authors and maintainers of this software the right to use, modify, distribute, sublicense, and otherwise deal with your contributions, including incorporating them into the software at their discretion. You also affirm that your contributions do not infringe on any third-party rights, and you have the necessary permissions to grant these rights.

Your contributions will be subject to the licensing terms determined by the authors and maintainers of this project. You acknowledge that the authors may choose to apply and/or change a license in the future that may differ from the current terms.

This software may include references or links to other open-source repositories or projects. Please note that we do not endorse, verify, or make any warranties regarding the reliability, accuracy, or suitability of these linked projects. You should use them at your own risk and discretion. Any issues, concerns, or liabilities arising from the use of these linked projects are separate from the responsibilities of the authors and maintainers of this software.

This software and its documentation may contain links to external websites or resources that are not maintained or controlled by the authors or maintainers of this project. We do not endorse, verify, or take responsibility for the content, accuracy, or availability of these external links. Clicking on such links is at your own risk, and any use of external websites or resources is subject to their respective terms and conditions.

Note

Maintenance of VoiceFixer 2 is powered by NeuralVox.

My newest audio-related AI project: TTS API - a open source Tortoise TTS API with streaming support that's coming soon. Join waitlist.

910882575/voicefixer2