Sanity check after NASH hackathon.

Yea it learns distortion nearly perfectly. This is essentially the same as the time align notebook, only with distortion

Installs and imports

%pip install -Uqq pip 

# Next line only executes on Colab. Colab users: Please enable GPU in Edit > Notebook settings
! [ -e /content ] && pip install -Uqq fastai git+https://github.com/drscotthawley/fastproaudio.git

# Additional installs for this tutorial
%pip install -q fastai_minima torchsummary pyzenodo3 wandb

# Install micro-tcn and auraloss packages (from source, will take a little while)
%pip install -q wheel --ignore-requires-python git+https://github.com/csteinmetz1/micro-tcn.git  git+https://github.com/csteinmetz1/auraloss

# After this cell finishes, restart the kernel and continue below

Note: you may need to restart the kernel to use updated packages.
Note: you may need to restart the kernel to use updated packages.
  WARNING: Missing build requirements in pyproject.toml for git+https://github.com/csteinmetz1/auraloss.
  WARNING: The project does not specify a build backend, and pip cannot fall back to setuptools without 'wheel'.
Note: you may need to restart the kernel to use updated packages.

from fastai.vision.all import *
from fastai.text.all import *
from fastai.callback.fp16 import *
import wandb
from fastai.callback.wandb import *
import torch
import torchaudio
import torchaudio.functional as F
import torchaudio.transforms as T
from IPython.display import Audio 
import matplotlib.pyplot as plt
import torchsummary
from fastproaudio.core import *
from pathlib import Path
import glob
import json
import re 

import warnings
# mel-spectrogram plot keeps throwing matplotlib deprecation warnings
warnings.filterwarnings( "ignore", module = "librosa\..*" ) 

Now Using the Real Dataset

We're going to use the multichannel version of the dataset. And we're going to stick to 16kHz to keep everything small and fast for this hackathon. We can upgrade to 48kHz later.

data_dir = '/home/shawley/datasets/pb_dist'  # Jacob made this, SHH fixed it up a bit
path = Path(data_dir) 

fnames_in = sorted(glob.glob(str(path)+'/*/input*'))
fnames_targ = sorted(glob.glob(str(path)+'/*/*targ*'))
ind = np.random.randint(len(fnames_in))   # pick one spot in the list of files
fnames_in[ind], fnames_targ[ind]

('/home/shawley/datasets/pb_dist/Train/input_164_.wav',
 '/home/shawley/datasets/pb_dist/Train/target_164__1__12.wav')

Input audio

input, sample_rate = torchaudio.load(fnames_in[ind])
print("sample_rate = ",sample_rate)
show_audio(input, sample_rate)

sample_rate =  44100
Shape: (1, 441000), Dtype: torch.float32, Duration: 10.0 s
Max:  0.583,  Min: -0.586, Mean:  0.000, Std Dev:  0.117