ECN>> genreml>> 返回
项目作者： adaros92

项目描述 ：
Implements a neural network model for predicting genres from visual and audio features extracted from audio files.
高级语言： Jupyter Notebook
项目地址: git://github.com/adaros92/genreml.git
创建时间： 2020-10-02T20:40:23Z
项目社区：https://github.com/adaros92/genreml
开源协议：MIT License
下载


CS467-Project

Repo for the CS467 capstone project

Contributing

Git

The usual flow will be as follows.

  1. Clone the repo
    1. git clone git@github.com:adaros92/CS467-Project.git
  2. Create new feature branch as feature/[name]
    1. git checkout main
    2. git pull
    3. git checkout -b feature/initial-structure
  3. Add code/make changes in feature branch
  4. Commit to remote
    1. git add .
    2. git commit -m "Add initial structure to repository and contributing section to readme.md"
    3. git push origin feature/initial-structure
  5. Submit a pull request at https://github.com/adaros92/CS467-Project/pulls from your feature branch to the main branch
  6. Resolve any comments locally and push a new revision to be reviewed
  7. Solve any merge conflicts and merge the pull request

Useful git commands and references.

Python Environment and Style

It’s recommended to use the Pycharm IDE for making contributions. You can get it for free here: https://www.jetbrains.com/pycharm/. This will help you catch style and syntax issues early.

We will try to follow PEP 8 whenever possible as documented here: https://www.python.org/dev/peps/pep-0008/. If you use Pycharm then you won’t need to read this because it will highlight issues for you :).

Installing Genreml

Install dependencies
Mac:

  1. brew install ffmpeg

Ubuntu/Debian:

  1. sudo apt-get install ffmpeg

To install the package from Pypi as documented in https://pypi.org/project/genreml run:

  1. pip3 install genreml
  2. or
  3. pip install genreml

To install the package from the repo with Unix-based OS run the following in CS467-Project directory

  1. bash install.sh

Test the installation by running the following in your terminal:

  1. genreml

If the installation fails, ensure you have pip installed and that pip is correctly mapped (it may be pip3 on your system). If the genreml command
doesn’t work then run the following as the entry point to the CLI.

  1. python3 -m model

Alternatively, you can travel to the src/model and run the following as the main entry point without installing the app

  1. python3 __main__.py

Deploying the Webapp

The webapp is being hosted on an us-east-2 EC2 instance in AWS account 146066720211. The public IPv4 address is 3.135.235.199
and the public IPv4 DNS is ec2-3-135-235-199.us-east-2.compute.amazonaws.com.

To deploy the webapp you will need to have SSH access to this instance. The key being used is available in
https://drive.google.com/file/d/1fc8yxqEZlNGF2s5lqqqSsr_oKaBHHcmA/view?usp=sharing to OSU students. To connect to the
instance you can run ssh -i “MyKey4.pem” ubuntu@ec2-3-135-235-199.us-east-2.compute.amazonaws.com from the same directory
as where MyKey4.pem is stored on your machine or provide a path to MyKey4.pem after -i argument like
ssh -i “/Users/adamsrosales/Downloads/MyKey4.pem” ubuntu@ec2-3-135-235-199.us-east-2.compute.amazonaws.com.

There is a shell script in this repo (deploy_webapp.sh) that copies the contents of the webapp to the right location on the
EC2 machine. You will need to provide the location of the private key on your machine when running it like so:
bash deploy_webapp.sh /Users/adamsrosales/Downloads/MyKey4.pem.

The deployment is configured like so.

In /etc/nginx/sites-enabled/ there’s a file called app with the following configuration which is the nginx
configuration for hosting the site.

  1. server {
  2.     listen 80;
  3.     server_name 3.135.235.199;
  5.     location / {
  6.         proxy_pass http://127.0.0.1:8000;
  7.         proxy_set_header Host $host;
  8.         proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
  9.     }
  10. }

If you make any changes run deploy_webapp.sh to copy the contents of the webapp to the server and run:

  1. sudo nginx -s reload
  2. sudo supervisorctl reload

To contents of the webapp are found in /home/ubuntu/webapp and they’re being run by gunicorn. To
see all the gunicorn processes run

  1. ps ax|grep gunicorn

To kill the gunicorn processes run

  1. sudo pkill gunicorn

To start the gunicorn process run this from the /home/ubuntu/webapp directory. This will start the site hosting.

  1. gunicorn -w 2 frontend:app

The site can be accessed by traveling to the public EC2 DNS name: 

  1. http://ec2-3-12-132-1.us-east-2.compute.amazonaws.com/

Running the CLI

The youtube functionality is not currently working due to youtube-dl takedown

Downloading audio files to your machine (This is no longer working due to youtube-dl takedown)

  1. genreml download -s "Black Dog" -a "Led Zeppelin"

You can do the same without installing the app by traveling to the model package and running

  1. python3 __main__.py download -s "Black Dog" -a "Led Zeppelin"

Processing audio files stored in some drive on your machine

  1. genreml process -fp "/Users/adamsrosales/Documents/audio-clips/"

You can do the same without installing the app by traveling to the model package and running

  1. python3 __main__.py process -fp "/Users/adamsrosales/Documents/audio-clips/"

The same can be run with a single file

  1. genreml process -fp "/Users/adamsrosales/Documents/audio-clips/Yummy_Justin Bieber_clip.wav"

To ignore certain features from the feature extraction process

  1. # Ignore spectrogram generation
  2. genreml process -fp "/Users/adamsrosales/Documents/audio-clips/fma_small/000" -e spectrogram

To change audio file type

  1. genreml process -fp "/Users/adamsrosales/Documents/audio-clips/fma_small/000" -af mp3

To change export color images and change image size to X in wide by Y in high

  1. genreml process -fp "/Users/adamsrosales/Documents/audio-clips/fma_small/000" -cmap None -fw 15.0 -fh 5.0

To predict genres from an audio file

  1. genreml classify -fp "/Users/adamsrosales/Documents/audio-clips/fma_small/000/000002.mp3"

To predict genres from a youtube URL

  1. genreml classify -yu "https://www.youtube.com/watch?v=Ui-_IUylvoA"

To predict genres from a compatible h5 model

  1. genreml classify -mp "/Users/adamsrosales/Downloads/FMA_model.h5" -yu "https://www.youtube.com/watch?v=Ui-_IUylvoA"

Using genreml as a Package in Python

You can import genreml after a successful installation like any Python package

  1. import genreml

See genreml/model/main for the hooks into the genreml functionality that the CLI has. Some examples below.

Run feature extraction on sample FMA MP3s packaged with application.

  1. from genreml.model.processing import audio
  3. audio_files = audio.AudioFiles()
  4. audio_files.extract_sample_fma_features()

Run feature extraction on already loaded data.

  1. from genreml.model.processing import audio
  3. audio_files = audio.AudioFiles()
  4. audio_files.extract_sample_fma_features()
  6. # Get the raw audio data from already loaded files
  7. audio_signal_data = []
  8. sample_rate_data = []
  9. for _, audio_obj in audio_files_processor.items():
  10.     audio_signal_data.append(audio_obj.audio_signal)
  11.     sample_rate_data.append(audio_obj.sample_rate)
  13. # Use that raw data for feature extraction with AudioData class
  14. audio_data_processor = AudioData(audio_signal_data, sample_rate_data)
  15. audio_data_processor.extract_features()

audio.AudioFiles() and AudioData() just create a dictionary of individual file paths to audio data.
You can extract the data from the collection as you would with any dictionary.

  1. audio_signals = []
  2. sample_rates = []
  3. for audio_key, object in audio_files.items():
  4.     audio_signals.append(object.audio_rate)
  5.     sample_rates.append(object.sample_rates)

You can also get all of the visual and non-visual features extracted from the audio collection itself.

  1. my_features = audio_files.features
  2. my_visual_features = audio_files.visual_features

Convert features to Pandas data frame

  1. df = audio_files.to_df()

Run feature extraction on a directory or filepath of your choice

  1. audio_files.extract_features("[YOUR_FILE_PATH]")

Run feature extraction on a directory or filepath of your choice but export results to a destination filepath

  1. audio_files.extract_features("[YOUR_FILE_PATH]", destination_filepath="[YOUR_DESTINATION_PATH]")

Change color of images generated from feature extraction

  1. audio_files.extract_features("[YOUR_FILE_PATH]", destination_filepath="[YOUR_DESTINATION_PATH]", cmap=None)

Change size of images generated to 15 by 5 inches

  1. audio_files.extract_features(
  2.     "[YOUR_FILE_PATH]", destination_filepath="[YOUR_DESTINATION_PATH]",
  3.     figure_width=15, figure_height=5
  4. )