Home

Awesome

🐧Pengi: An Audio Language Model for Audio Tasks

[Paper] [Checkpoints]

Pengi is an Audio Language Model that leverages Transfer Learning by framing all audio tasks as text-generation tasks. It takes as input, an audio recording, and text, and generates free-form text as output. The unified architecture of Pengi enables open-ended tasks and close-ended tasks without any additional fine-tuning or task-specific extensions. image

News

[Sep 23] 🐧Pengi is accepted at NeurIPS 2023

Setup

  1. You are required to install the dependencies: pip install -r requirements.txt. If you have conda installed, you can run the following:
cd Pengi && \
conda create -n pengi python=3.8 && \
conda activate pengi && \
pip install -r requirements.txt
  1. Download Pengi weights: Pretrained Model [Zenodo]
  2. Move the base.pth and base_no_text_enc.pth under configs folder

Supported models

pengi_1-main_arch The wrapper supports two models. The base option is Pengi architecture reported in paper and shown above. The base_no_text_enc is the Pengi architecture without the text encoder and only $m_2$ to encode tokenized text. All models only support 44.1 kHz input audio.

Usage

The wrapper provides an easy way to get Pengi output given and audio and text input. To use the wrapper, inputs required are:

Supported functions:

Text generation

from wrapper import PengiWrapper as Pengi

pengi = Pengi(config="<choice of config>")

generated_response = pengi.generate(audio_paths=audio_file_paths,
                                    text_prompts=["generate metadata"], 
                                    add_texts=[""], 
                                    max_len=30, 
                                    beam_size=3, 
                                    temperature=1.0, 
                                    stop_token=' <|endoftext|>'
                                    )

Audio description

from wrapper import PengiWrapper as Pengi

pengi = Pengi(config="<choice of config>")

generated_summary = pengi.describe(audio_paths=audio_file_paths,
                                    max_len=30, 
                                    beam_size=3, 
                                    temperature=1.0, 
                                    stop_token=' <|endoftext|>'
                                    )

Generate audio, audio prefix and prompt embeddings

audio_prefix, audio_embeddings = pengi.get_audio_embeddings(audio_paths=audio_file_paths)

text_prefix, text_embeddings = pengi.get_prompt_embeddings(prompts=["generate metadata"])

Citation

@inproceedings{deshmukh2023pengi,
 author = {Deshmukh, Soham and Elizalde, Benjamin and Singh, Rita and Wang, Huaming},
 booktitle = {Advances in Neural Information Processing Systems},
 editor = {A. Oh and T. Neumann and A. Globerson and K. Saenko and M. Hardt and S. Levine},
 pages = {18090--18108},
 publisher = {Curran Associates, Inc.},
 title = {Pengi: An Audio Language Model for Audio Tasks},
 url = {https://proceedings.neurips.cc/paper_files/paper/2023/file/3a2e5889b4bbef997ddb13b55d5acf77-Paper-Conference.pdf},
 volume = {36},
 year = {2023}
}

Contributing

This project welcomes contributions and suggestions. Most contributions require you to agree to a Contributor License Agreement (CLA) declaring that you have the right to, and actually do, grant us the rights to use your contribution. For details, visit https://cla.opensource.microsoft.com.

When you submit a pull request, a CLA bot will automatically determine whether you need to provide a CLA and decorate the PR appropriately (e.g., status check, comment). Simply follow the instructions provided by the bot. You will only need to do this once across all repos using our CLA.

This project has adopted the Microsoft Open Source Code of Conduct. For more information see the Code of Conduct FAQ or contact opencode@microsoft.com with any additional questions or comments.

Trademarks

This project may contain trademarks or logos for projects, products, or services. Authorized use of Microsoft trademarks or logos is subject to and must follow Microsoft's Trademark & Brand Guidelines. Use of Microsoft trademarks or logos in modified versions of this project must not cause confusion or imply Microsoft sponsorship. Any use of third-party trademarks or logos are subject to those third-party's policies.