Melody-Lyric-Generator is designed to combine melody generation and lyric generation for creating cohesive musical content. It processes, analyzes, and generates melodies using MIDI data, while simultaneously handling lyric extraction and fine-tuning for generating Spanish-language text. This project employs advanced datasets, custom preprocessing functions, and machine learning models to produce high-quality musical and lyrical outputs.
Content:
Lyrics Generation processes and analyzes song lyrics using the LyricsGenius API, cleaning the data and identifying emotional content through a lexicon-based approach. Additionally, it fine-tunes the pre-trained datificate/gpt2-small-spanish model for text generation.
Key Features:
- Data Cleaning: Removes unnecessary elements such as metadata, punctuation, and stopwords.
- Emotional Analysis: Identifies emotions in lyrics using a specialized lexicon.
- Fine-Tuning GPT: Trains the model across multiple datasets, optimizing for coherent and emotionally rich text.
Melody Generation processes, analyzes, and generates melodies using the MAESTRO dataset and custom MIDI handling functions. A machine learning model is trained to predict musical notes, evaluate performance, and produce cohesive melodies.
Key Features:
- Custom MIDI Functions: Includes tools for note extraction, sequence creation, and MIDI playback.
- Neural Network Architecture: Trains on features like pitch, step, and duration to predict melodies.
- Evaluation Metrics: Uses RMSE and R² to measure the quality of generated melodies.