PyTorch wrappers for using your model in audacity!

Overview

audacitorch

This package contains utilities for prepping PyTorch audio models for use in Audacity. More specifically, it provides abstract classes for you to wrap your waveform-to-waveform and waveform-to-labels models (see the Deep Learning for Audacity website to learn more about deep learning models for audacity).

Table of Contents


img

Download Audacity with Deep Learning

Our work has not yet been merged to the main build of Audacity, though it will be soon. You can keep track of its progress by viewing our pull request. In the meantime, you can download an alpha version of Audacity + Deep Learning here.

Installing

You can install audacitorch using pip:

pip install -e "git+https://github.com/hugofloresgarcia/audacitorch.git#egg=audacitorch"

Contributing Models to Audacity

Supported Torch versions

audacitorch requires for your model to be able to run in Torch 1.9.0, as that's what the Audacity torchscript interpreter uses.

Deep Learning Effect and Analyzer

Audacity is equipped with a wrapper framework for deep learning models written in PyTorch. Audacity contains two deep learning tools: Deep Learning Effect and Deep Learning Analyzer.
Deep Learning Effect performs waveform to waveform processing, and is useful for audio-in-audio-out tasks (such as source separation, voice conversion, style transfer, amplifier emulation, etc.), while Deep Learning Analyzer performs waveform to labels processing, and is useful for annotation tasks (such as sound event detection, musical instrument recognition, automatic speech recognition, etc.). audacitorch contains two abstract classes for serializing two types of models: waveform-to-waveform and waveform-to-labels. The classes are WaveformToWaveformBase, and WaveformToLabelsBase, respectively.

Choosing an Effect Type

Waveform to Waveform models

As shown in the effect diagram, Waveform-to-waveform models receive a single multichannel audio track as input, and may write to a variable number of new audio tracks as output.

Example models for waveform-to-waveform effects include source separation, neural upsampling, guitar amplifier emulation, generative models, etc. Output tensors for waveform-to-waveform models must be multichannel waveform tensors with shape (num_output_channels, num_samples). For every audio waveform in the output tensor, a new audio track is created in the Audacity project.

Waveform to Labels models

As shown in the effect diagram, Waveform-to-labels models receive a single multichannel audio track as input, and may write to an output label track as output. The waveform-to-labels effect can be used for many audio analysis applications, such as voice activity detection, sound event detection, musical instrument recognition, automatic speech recognition, etc. The output for waveform-to-labels models must be a tuple of two tensors. The first tensor corresponds to the class indexes for each label present in the waveform, shape (num_timesteps,). The second tensor must contain timestamps with start and stop times for each label, shape (num_timesteps, 2).

What If My Model Uses a Spectrogram as Input/Output?

If your model uses a spectrogram as input/output, you'll need to wrap your forward pass with some torchscript-compatible preprocessing/postprocessing. We recommend using torchaudio, writing your own preprocessing transforms in their own nn.Module, or writing your PyTorch-only preprocessing and placing it in WaveformToWaveform.do_forward_pass or WaveformToLabels.do_forward_pass. See the compatibility section for more info.

Model Metadata

Certain details about the model, such as its sample rate, tool type (e.g. waveform-to-waveform or waveform-to-labels), list of labels, etc. must be provided by the model contributor in a separate metadata.json file. In order to help users choose the correct model for their required task, model contributors are asked to provide a short and long description of the model, the target domain of the model (e.g. speech, music, environmental, etc.), as well as a list of tags or keywords as part of the metadata. See here for an example metadata dictionary.

Metadata Spec

required fields:

  • sample_rate (int)
    • range (0, 396000)
    • Model sample rate. Input tracks will be resampled to this value.
  • domains (List[str])
    • List of data domains for the model. The list should contain any of the following strings (any others will be ignored): ["music", "speech", "environmental", "other"]
  • short_description(str)
    • max 60 chars
    • short description of the model. should contain a brief message with the model's purpose, e.g. "Use me for separating vocals from the background!".
  • long_description (str)
    • max 280 chars
    • long description of the model. Shown in the detailed view of the model UI.
  • tags (List[str])
    • list of tags (to be shown in the detailed view)
    • each tag should be 15 characters max
    • max 5 tags per model.
  • labels (List[str)
    • output labels for the model. Depending on the effect type, this field means different things
    • waveform-to-waveform
      • name of each output source (e.g. drums, bass, vocal). To create the track name for each output source, each one of the labels will be appended to the mixture track's name.
    • waveform-to-labels:
      • This should be classlist for model. The class indexes output by the model during a forward pass will be used to index into this classlist.
  • effect_type (str)
    • Target effect for this model. Must be one of ["waveform-to-waveform", "waveform-to-labels"].
  • multichannel (bool)
    • If multichannel is set to true, stereo tracks are passed to the model as multichannel audio tensors, with shape (2, n). Note that this means that the input could either be a mono track with shape (1, n) or stereo track with shape (2, n).
    • If multichannel is set to false, stereo tracks are downmixed, meaning that the input audio tensor will always be shape (1, n).

Making Your Model Built-In To Audacity

By default, users have to click on the Add From HuggingFace button on the Audacity Model Manager and enter the desired repo's ID to install a community contributed model. If you, instead, would like your community contributed model to show up in Audacity's Model Manager by default, please open a request here.

Example - Waveform-to-Waveform model

Here's a minimal example for a model that simply boosts volume by multiplying the incoming audio by a factor of 2.

We can sum up the whole process into 4 steps:

  1. Developing your model
  2. Wrapping your model using audacitorch
  3. Creating a metadata document
  4. Exporting to HuggingFace

Developing your model

First, we create our model. There are no internal constraints on what the internal model architecture should be, as long as you can use torch.jit.script or torch.jit.trace to serialize it, and it is able to meet the input-output constraints specified in waveform-to-waveform and waveform-to-labels models.

import torch
import torch.nn as nn

class MyVolumeModel(nn.Module):

    def forward(self, x: torch.Tensor) -> torch.Tensor:
        # do the neural net magic!
        x = x * 2

        return x

Making sure your model is compatible with torchscript

PyTorch makes it really easy to deploy your Python models in C++ by using torchscript, an intermediate representation format for torch models that can be called in C++. Many of Python's built-in functions are supported by torchscript. However, not all Python operations are supported by the torchscript environment, meaning that you are only allowed to use a subset of Python operations in your model code. See the torch.jit docs to learn more about writing torchscript-compatible code.

If your model computes spectrograms (or requires any kind of preprocessing/postprocessing), make sure those operations are compatible with torchscript, like torchaudio's operation set.

Useful links:

Wrapping your model using audacitorch

Now, we create a wrapper class for our model. Because our model returns an audio waveform as output, we'll use WaveformToWaveformBase as our parent class. For both WaveformToWaveformBase and WaveformToLabelsBase, we need to implement the do_forward_pass method with our processing code. See the docstrings for more details.

from audacitorch import WaveformToWaveformBase

class MyVolumeModelWrapper(WaveformToWaveformBase):
    
    def do_forward_pass(self, x: torch.Tensor) -> torch.Tensor:
        
        # do any preprocessing here! 
        # expect x to be a waveform tensor with shape (n_channels, n_samples)

        output = self.model(x)

        # do any postprocessing here!
        # the return value should be a multichannel waveform tensor with shape (n_channels, n_samples)
    
        return output

Creating a metadata document

Audacity models need a metadata file. See the metadata spec to learn about the required fields.

metadata = {
    'sample_rate': 48000, 
    'domain_tags': ['music', 'speech', 'environmental'],
    'short_description': 'Use me to boost volume by 3dB :).',
    'long_description':  'This description can be a max of 280 characters aaaaaaaaaaaaaaaaaaaa.',
    'tags': ['volume boost'],
    'labels': ['boosted'],
    'effect_type': 'waveform-to-waveform',
    'multichannel': False,
}

All set! We can now proceed to serialize the model to torchscript and save the model, along with its metadata.

from pathlib import Path
from audacitorch.utils import save_model, validate_metadata, \
                              get_example_inputs, test_run

# create a root dir for our model
root = Path('booster-net')
root.mkdir(exist_ok=True, parents=True)

# get our model
model = MyVolumeModel()

# wrap it
wrapper = MyVolumeModelWrapper(model)

# serialize it using torch.jit.script, torch.jit.trace,
# or a combination of both. 

# option 1: torch.jit.script 
# using torch.jit.script is preferred for most cases, 
# but may require changing a lot of source code
serialized_model = torch.jit.script(wrapper)

# option 2: torch.jit.trace
# using torch.jit.trace is typically easier, but you
# need to be extra careful that your serialized model behaves 
# properly after tracing
example_inputs = get_example_inputs()
serialized_model = torch.jit.trace(wrapper, example_inputs[0], 
                                    check_inputs=example_inputs)

# take your model for a test run!
test_run(serialized_model)

# check that we created our metadata correctly
success, msg = validate_metadata(metadata)
assert success

# save!
save_model(serialized_model, metadata, root)

Exporting to HuggingFace

You should now have a directory structure that looks like this:

/booster-net/
/booster-net/model.pt
/booster-net/metadata.json

This will be the repository for your audacity model. Make sure to add a readme with the audacity tag in the YAML metadata, so it show up on the explore tab of Audacity's Deep Learning Tools.

Create a README.md inside booster-net/, and add the following header:

in README.md

---
tags: audacity
---

Awesome! It's time to push to HuggingFace. See their documentation for adding a model to the HuggingFace model hub.

Debugging Your Model in Audacity

After serializing, you may need to debug your model inside Audacity, to make sure that it handles inputs correctly, doesn't crash while processing, and produces the correct output. While debugging, make sure your model isn't available through other users through the Explore HuggingFace button by temporarily removing the audacity tag from your README file. If your model fails internally while processing audio, you may see something like this:

To debug, you can access the error logs through the Help menu, in Help->Diagnostics->Show Log.... Any torchscript errors that may occur during the forward pass will be redirected here.

Example - Exporting a Pretrained Asteroid model

See this example notebook, where we serialize a pretrained ConvTasNet model for speech separation using the Asteroid source separation library.

Example - Exporting a Pretrained S2T model

See this example notebook, where we serialize a pretrained speech to text transformer from Facebook.


You might also like...
Convert Pytorch model to onnx or tflite, and the converted model can be visualized by Netron
Convert Pytorch model to onnx or tflite, and the converted model can be visualized by Netron

Convert Pytorch model to onnx or tflite, and the converted model can be visualized by Netron

Count the MACs / FLOPs of your PyTorch model.

THOP: PyTorch-OpCounter How to install pip install thop (now continously intergrated on Github actions) OR pip install --upgrade git+https://github.co

Objective of the repository is to learn and build machine learning models using Pytorch. 30DaysofML Using Pytorch
Objective of the repository is to learn and build machine learning models using Pytorch. 30DaysofML Using Pytorch

30 Days Of Machine Learning Using Pytorch Objective of the repository is to learn and build machine learning models using Pytorch. List of Algorithms

Python scripts form performing stereo depth estimation using the high res stereo model in PyTorch .
Python scripts form performing stereo depth estimation using the high res stereo model in PyTorch .

PyTorch-High-Res-Stereo-Depth-Estimation Python scripts form performing stereo depth estimation using the high res stereo model in PyTorch. Stereo dep

PyTorch implementation of CVPR 2020 paper (Reference-Based Sketch Image Colorization using Augmented-Self Reference and Dense Semantic Correspondence) and pre-trained model on ImageNet dataset

Reference-Based-Sketch-Image-Colorization-ImageNet This is a PyTorch implementation of CVPR 2020 paper (Reference-Based Sketch Image Colorization usin

Deploy pytorch classification model using Flask and Streamlit

Deploy pytorch classification model using Flask and Streamlit

Pytorch implementation of 'Fingerprint Presentation Attack Detector Using Global-Local Model'
Pytorch implementation of 'Fingerprint Presentation Attack Detector Using Global-Local Model'

RTK-PAD This is an official pytorch implementation of 'Fingerprint Presentation Attack Detector Using Global-Local Model', which is accepted by IEEE T

Model search is a framework that implements AutoML algorithms for model architecture search at scale
Model search is a framework that implements AutoML algorithms for model architecture search at scale

Model search (MS) is a framework that implements AutoML algorithms for model architecture search at scale. It aims to help researchers speed up their exploration process for finding the right model architecture for their classification problems (i.e., DNNs with different types of layers).

Implementation of STAM (Space Time Attention Model), a pure and simple attention model that reaches SOTA for video classification
Implementation of STAM (Space Time Attention Model), a pure and simple attention model that reaches SOTA for video classification

STAM - Pytorch Implementation of STAM (Space Time Attention Model), yet another pure and simple SOTA attention model that bests all previous models in

Comments
  • Audacitorch install prevents import of torchaudio

    Audacitorch install prevents import of torchaudio

    I've trained a model that acts on MFCCs I get using torchaudio. When I installed audacitorch to wrap my model (in Colab), it prevents me from importing torchaudio with the message OSError: libtorch_cuda_cpp.so: cannot open shared object file: No such file or directory. It looks like installing audacitorch uninstalls and re-installs torch - might that have something to do with it?

    opened by jonesmo 3
  • Labeler example revised

    Labeler example revised

    Added a model more compatible with torchscript, removed unneeded sections of the notebook, and a small aside on where to get more information about the huggingface transformer module's compatibility with torchscript

    opened by aldo-aguilar 0
  • tweaks to labeler_example.ipynb

    tweaks to labeler_example.ipynb

    Hey Aldo,

    Good job getting the wavtogether. I just have a couple of small suggested edits. The first one is to add a section at the end that points them to how to upload a model to HuggingFace. The second boils down to adding a couple of hyperlinks to give information. I'll provide examples where it would be cool to add a link below.

    In "Audacity WaveformToLabels Example" In this notebook we will load in [a speech to text model](hyperlink to model)

    In "Wraping the model" torchaudacity provides a WaveformToLabels class . We will use this as a base class for our pretrained models wrapper. The WaveformToLabels class provides us with tests to ensure that our model is receiving properly sized input, and outputting the expected tensor shapes for Audacity's [Deep Learning Analyzer](hyperlink or explanation).

    In "Saving Our Model & Metadata' We will now save the wrapped model locally by tracing it with torchscript, and generating a ScriptModule or ScriptFunction using torch.jit.script. We can then use torchaudacity's utility function save_model to save the model and meta data easily.

    opened by bryan-pardo 0
Owner
Hugo Flores García
PhD @interactiveaudiolab
Hugo Flores García
OptaPlanner wrappers for Python. Currently significantly slower than OptaPlanner in Java or Kotlin.

OptaPy is an AI constraint solver for Python to optimize the Vehicle Routing Problem, Employee Rostering, Maintenance Scheduling, Task Assignment, School Timetabling, Cloud Optimization, Conference Scheduling, Job Shop Scheduling, Bin Packing and many more planning problems.

OptaPy 190 Sep 19, 2022
Python wrappers to the C++ library SymEngine, a fast C++ symbolic manipulation library.

SymEngine Python Wrappers Python wrappers to the C++ library SymEngine, a fast C++ symbolic manipulation library. Installation Pip See License section

null 130 Sep 29, 2022
In this project we investigate the performance of the SetCon model on realistic video footage. Therefore, we implemented the model in PyTorch and tested the model on two example videos.

Contrastive Learning of Object Representations Supervisor: Prof. Dr. Gemma Roig Institutions: Goethe University CVAI - Computational Vision & Artifici

Dirk Neuhäuser 5 Feb 27, 2022
Step by Step on how to create an vision recognition model using LOBE.ai, export the model and run the model in an Azure Function

Step by Step on how to create an vision recognition model using LOBE.ai, export the model and run the model in an Azure Function

El Bruno 3 Mar 30, 2022
Cancer-and-Tumor-Detection-Using-Inception-model - In this repo i am gonna show you how i did cancer/tumor detection in lungs using deep neural networks, specifically here the Inception model by google.

Cancer-and-Tumor-Detection-Using-Inception-model In this repo i am gonna show you how i did cancer/tumor detection in lungs using deep neural networks

Deepak Nandwani 1 Jan 1, 2022
Capture all information throughout your model's development in a reproducible way and tie results directly to the model code!

Rubicon Purpose Rubicon is a data science tool that captures and stores model training and execution information, like parameters and outcomes, in a r

Capital One 80 Sep 29, 2022
MBPO (paper: When to trust your model: Model-based policy optimization) in offline RL settings

offline-MBPO This repository contains the code of a version of model-based RL algorithm MBPO, which is modified to perform in offline RL settings Pape

LxzGordon 1 Oct 24, 2021
😇A pyTorch implementation of the DeepMoji model: state-of-the-art deep learning model for analyzing sentiment, emotion, sarcasm etc

------ Update September 2018 ------ It's been a year since TorchMoji and DeepMoji were released. We're trying to understand how it's being used such t

Hugging Face 858 Sep 18, 2022