Skip to content

Commit

Permalink
feat: added ljspeech, youtube, and transforms
Browse files Browse the repository at this point in the history
  • Loading branch information
flavioschneider committed Jul 25, 2022
0 parents commit 39993db
Showing 12 changed files with 559 additions and 0 deletions.
39 changes: 39 additions & 0 deletions .github/workflows/python-publish.yml
Original file line number Diff line number Diff line change
@@ -0,0 +1,39 @@
# This workflow will upload a Python Package using Twine when a release is created
# For more information see: https://help.github.com/en/actions/language-and-framework-guides/using-python-with-github-actions#publishing-to-package-registries

# This workflow uses actions that are not certified by GitHub.
# They are provided by a third-party and are governed by
# separate terms of service, privacy policy, and support
# documentation.

name: Upload Python Package

on:
release:
types: [published]

permissions:
contents: read

jobs:
deploy:

runs-on: ubuntu-latest

steps:
- uses: actions/checkout@v3
- name: Set up Python
uses: actions/setup-python@v3
with:
python-version: '3.x'
- name: Install dependencies
run: |
python -m pip install --upgrade pip
pip install build
- name: Build package
run: python -m build
- name: Publish package
uses: pypa/gh-action-pypi-publish@27b31702a0e7fc50959f5ad993c78deac1bdfc29
with:
user: __token__
password: ${{ secrets.PYPI_API_TOKEN }}
2 changes: 2 additions & 0 deletions .gitignore
Original file line number Diff line number Diff line change
@@ -0,0 +1,2 @@
__pycache__
.mypy_cache
40 changes: 40 additions & 0 deletions .pre-commit-config.yaml
Original file line number Diff line number Diff line change
@@ -0,0 +1,40 @@
repos:
- repo: https://github.com/pre-commit/pre-commit-hooks
rev: v2.3.0
hooks:
- id: end-of-file-fixer
- id: trailing-whitespace

# Formats code correctly
- repo: https://github.com/psf/black
rev: 21.12b0
hooks:
- id: black
args: [
'--experimental-string-processing'
]

# Sorts imports
- repo: https://github.com/pycqa/isort
rev: 5.10.1
hooks:
- id: isort
name: isort (python)

# Checks unused imports, like lengths, etc
- repo: https://gitlab.com/pycqa/flake8
rev: 4.0.0
hooks:
- id: flake8
args: [
'--per-file-ignores=__init__.py:F401',
'--max-line-length=88',
'--ignore=E203'
]

# Checks types
- repo: https://github.com/pre-commit/mirrors-mypy
rev: 'v0.971'
hooks:
- id: mypy
additional_dependencies: [data-science-types>=0.2, torch>=1.6]
21 changes: 21 additions & 0 deletions LICENSE
Original file line number Diff line number Diff line change
@@ -0,0 +1,21 @@
MIT License

Copyright (c) 2022 archinet.ai

Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.
98 changes: 98 additions & 0 deletions README.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,98 @@

# Audio Data - PyTorch

A collection of useful audio datasets and transforms for PyTorch.

## Install

```bash
pip install audio-data-pytorch
```

[![PyPI - Python Version](https://img.shields.io/pypi/v/audio-data-pytorch?style=flat&colorA=0f0f0f&colorB=0f0f0f)](https://pypi.org/project/audio-data-pytorch/)

## Datasets

### WAV Dataset

Load one or multiple folders of `.wav` files as dataset.

```py
from audio_data_pytorch import WAVDataset

dataset = WAVDataset(path=['my/path1', 'my/path2'])
```

#### Full API:
```py
WAVDataset(
path: Union[str, Sequence[str]], # Path or list of paths from which to load files
recursive: bool = False # Recursively load files from provided paths
with_sample_rate: bool = False, # Returns sample rate as second argument
transforms: Optional[Callable] = None, # Transforms to apply to audio files
)
```

### LJSpeech Dataset
An unsupervised dataset for LJSpeech with voice only data
```py
from audio_data_pytorch import LJSpeechDataset

dataset = LJSpeechDataset(root='./data')

dataset[0] # (1, 158621)
dataset[1] # (1, 153757)
```

#### Full API:
```py
LJSpeechDataset(
root: str = "./data", # The root where the dataset will be downloaded
with_sample_rate: bool = False, # Returns sample rate as second argument
transforms: Optional[Callable] = None, # Transforms to apply to audio files
)
```

### Youtube Dataset
A wrapper around yt-dlp that automatically downloads the audio source of Youtube videos.

```py
dataset = YoutubeDataset(
root='./data',
urls=[
"https://www.youtube.com/watch?v=dQw4w9WgXcQ",
"https://www.youtube.com/watch?v=BZ-_KQezKmU",
],
crop_length=10 # Crop source in 10s chunks (optional but suggested)
)
dataset[0] # (2, 480000)
```

#### Full API:
```py
dataset = YoutubeDataset(
urls: Sequence[str], # The list of youtube urls
root: str = "./data", # The root where the dataset will be downloaded
crop_length: Optional[int] = None, # Crops the source into chunks of `crop_length` seconds
with_sample_rate: bool = False, # Returns sample rate as second argument
transforms: Optional[Callable] = None, # Transforms to apply to audio files
)
```


## Transforms

An example

```py

crop = Crop(22050) # Crop start of audio track

transforms = nn.Sequential(
Resample(source=48000, target=22050), # Resample from 48kHz to 22kHz
OverlapChannels(), # Overap channels by sum (C, N) -> (1, N)
RandomCrop(22050 * 3), # Random crop from file
Scale(0.8) # Scale waveform
)

```
4 changes: 4 additions & 0 deletions audio_data_pytorch/__init__.py
Original file line number Diff line number Diff line change
@@ -0,0 +1,4 @@
from .ljspeech_dataset import LJSpeechDataset
from .transforms import Crop, OverlapChannels, RandomCrop, Resample, Scale
from .wav_dataset import WAVDataset
from .youtube_dataset import YoutubeDataset
57 changes: 57 additions & 0 deletions audio_data_pytorch/ljspeech_dataset.py
Original file line number Diff line number Diff line change
@@ -0,0 +1,57 @@
import os
import tarfile

import requests # type: ignore
from tqdm import tqdm

from .utils import camel_to_snake
from .wav_dataset import WAVDataset


class LJSpeechDataset(WAVDataset):

data_url = "https://data.keithito.com/data/speech/LJSpeech-1.1.tar.bz2"
data_tar_file = "LJSpeech-1.1.tar.bz2"
data_waws_path = "LJSpeech-1.1/wavs"

def __init__(self, root: str = "./data", **kwargs) -> None:
self.root = root

if not os.path.exists(self.data_path):
print(
f"Data not found in {self.data_path}, downloading {self.data_tar_file}"
)
self.download()

super().__init__(path=self.wavs_path, **kwargs)

@property
def data_path(self) -> str:
return os.path.join(self.root, camel_to_snake(self.__class__.__name__))

@property
def file_path(self) -> str:
return os.path.join(self.data_path, self.data_tar_file)

@property
def wavs_path(self) -> str:
return os.path.join(self.data_path, self.data_waws_path)

def download(self) -> None:
os.makedirs(self.data_path, exist_ok=True)
response = requests.get(self.data_url, stream=True)
block_size = 1024 # Kibibyte
progress_bar = tqdm(total=block_size, unit="iB", unit_scale=True)

with open(self.file_path, "wb") as file:
for data in response.iter_content(block_size):
progress_bar.update(len(data))
file.write(data)
progress_bar.close()
self.decompress()

def decompress(self) -> None:
print(f"Decompressing {self.data_tar_file} to {self.data_path}")
file = tarfile.open(self.file_path)
file.extractall(self.data_path)
file.close()
82 changes: 82 additions & 0 deletions audio_data_pytorch/transforms.py
Original file line number Diff line number Diff line change
@@ -0,0 +1,82 @@
import random

import torch
import torchaudio
from torch import Tensor, nn


class Crop(nn.Module):
"""Crops waveform to fixed size"""

def __init__(self, size: int, start: int = 0) -> None:
super().__init__()
self.size = size
self.start = start

def forward(self, x: Tensor) -> Tensor:
x = x[:, self.start :]
channels, length = x.shape

if length < self.size:
padding_length = self.size - length
padding = torch.zeros(channels, padding_length).to(x)
return torch.cat([x, padding], dim=1)
else:
return x[:, 0 : self.size]


class RandomCrop(nn.Module):
"""Crops random chunk from the waveform"""

def __init__(self, size: int) -> None:
super().__init__()
self.size = size

def forward(self, x: Tensor) -> Tensor:
# Pick start position
length = x.shape[1]
start = random.randint(0, max(length - self.size, 0))
# Crop from random start
x = x[:, start:]
channels, length = x.shape
# Pad to end if not large enough, else crop end
if length < self.size:
padding_length = self.size - length
padding = torch.zeros(channels, padding_length).to(x)
return torch.cat([x, padding], dim=1)
else:
return x[:, 0 : self.size]


class OverlapChannels(nn.Module):
"""Overlaps all channels into one"""

def forward(self, x: Tensor) -> Tensor:
return torch.sum(x, dim=0, keepdim=True) # 'c l -> 1 l'


class Resample(nn.Module):
"""Resamples frequency of waveform"""

def __init__(self, source: int, target: int):
super().__init__()
self.transform = torchaudio.transforms.Resample(
orig_freq=source, new_freq=target
)

def forward(self, x: Tensor) -> Tensor:
return self.transform(x)


class Scale(nn.Module):
"""Scales waveform (change volume)"""

def __init__(
self,
scale: float,
):
super().__init__()
self.scale = scale

def forward(self, x: Tensor) -> Tensor:
return x * self.scale
15 changes: 15 additions & 0 deletions audio_data_pytorch/utils.py
Original file line number Diff line number Diff line change
@@ -0,0 +1,15 @@
import re
from typing import Optional, TypeVar

from typing_extensions import TypeGuard

T = TypeVar("T")


def exists(val: Optional[T]) -> TypeGuard[T]:
return val is not None


def camel_to_snake(name: str) -> str:
name = re.sub("(.)([A-Z][a-z]+)", r"\1_\2", name)
return re.sub("([a-z0-9])([A-Z])", r"\1_\2", name).lower()
Loading

0 comments on commit 39993db

Please sign in to comment.