This project predicts which genres a song will do the best in given its lyrics and a one-word theme. The theme is generated with OpenAI's GPT 4.0 Mini based on the song's lyrics.
The project is split into two parts: the scraper and the model. The scraper pulls song information (~2400 songs) from Spotify across 18 different genres, and the model uses this information to predict which genres work best for song lyrics that a user inputs. The scraped songs are stored in a local MongoDB database.
To isolate project dependencies, create a virtual environment:
python3 -m venv venv
Activate the virtual environment with the following command:
source ./venv/bin/activate
Install all necessary libraries by running:
pip install -r requirements.txt
Make sure MongoDB is intalled and running on your computer. You can do so by following the instructions on the Mongo website: https://www.mongodb.com/docs/manual/installation/
There are two third-party services that the scraper depends on: 2Captcha for solving the reCaptcha v3 for Spotify's login and OpenAI for generating the one-word topics for each song.
First, go to https://2captcha.com/ and sign into your account. If you don't have an account, you'll have to create one and load funds to use the autonomous captcha solving. After logging in, copy your API key on the home page and either set an environment variable TWOCAP_KEY to the key, or set the constant TWOCAP_KEY in scraper/config.py to the key.
Next, head to https://platform.openai.com/ and sign into / create an account. You'll need to load funds to generate one-word song topics. There are three pieces of information you'll need: your organization ID, your project ID, and a secret API key for your project. To get your organization ID, go to Organization > General. To get your project ID, go to Project > General. To create a secret key, go to Project > API keys and follow the instructions.
You can either set these environment variables: OPENAI_ORGANIZATION_ID for organization ID, OPENAI_PROJECT_ID for project ID, and OPENAI_API_KEY for the secret key OR navigate to scraper/topic_classifier.py and change the arguments for the OpenAI client object.
Finally, change the placeholder account credentials to those of your Spotify account in the SPOTIFY_ACCOUNT tuple of scraper/config.py.
Make sure your Mongo server is running locally on port 27017. Then, run the main.py file in the scraper directory.
This will take ~30 minutes to run until completion. You can toggle the variable CLEAR_DATABASE_ON_RESTART in scraper/config.py to reset the MongoDB database on each run of the scraper.
Find song lyrics that you want to analyze and put them into lyrics.txt. Then, run the main.py file in the model directory, but make sure you are using an environment/IDE that treats the scraper directory as a library. This is because the model uses generate_topic from the scraper to generate a topic for the song input. As an alternative, you can comment out the from scraper.topic_classifier import generate_topic line in model/main.py and replace this block of code:
topic, err = generate_topic(lyrics_text)
if err:
print(f"Could not generate song lyrics: {err}")
sys.exit()
With a hardcoded topic like this:
topic = "Love"
The output should look something like this:
Getting songs DataFrame from MongoDB...
Training the model...
Pulling the test song lyrics from lyrics.txt...
Generating topic from the song lyrics...
Song topic: Embarrassment
Predicting the success of a sample song...
Here are the probabilities of this song's success in each genre:
Funk: 14.00%
K-Pop: 47.00%
Folk/Americana: 41.00%
Reggaeton: 3.00%
Indie: 32.00%
Pop: 71.00%
Latin: 22.00%
Kids: 6.00%
Hip-Hop: 22.00%
R&B: 42.00%
Jazz: 40.62%
Blues: 10.00%
Country: 61.00%
Metal: 31.00%
Christian/Gospel: 46.28%
Rock: 53.73%
Dance/Electronic: 42.00%
Showing model visualizations...
Done!