QaraamiGen AI
From Oral Heritage to Digital Music Systems
The Challenge: Digital Extinction
Somali music, particularly the Qaraami tradition, relies on complex pentatonic scales and microtonal nuances that Western MIDI standards cannot capture. It is an oral tradition, largely undocumented in sheet music.
As the masters of this tradition pass away, we risk losing centuries of cultural heritage. Current AI models are trained on Western datasets. They cannot generate, understand, or preserve the soul of Somali music.
System Architecture
Data Collection
Encoding raw audio from Oud recordings into custom symbolic representations.
Scale Analysis
DSP algorithms to detect and map microtonal Qaraami intervals.
Model Training
Experimenting with LSTMs and Transformers to learn rhythmic probabilities.
Generation
Synthesizing new melodies that respect traditional structural rules.
Current Progress: AI Transcription Engine
I have successfully built an end-to-end MIR pipeline in Python that ingests raw audio and outputs standard MusicXML.
- Source Separation (HPSS)
- pYIN Pitch Detection
- Automated Quantization
Inference Output: Raw .musicxml Generated via Python
Research Trajectory
Short TermComplete the first annotated dataset of Somali Oud Scales (MIDI + Audio).
At Graduate StudyLeverage advanced resources to move from statistical models to Deep Learning architectures that handle microtonal improvisation.
Long TermBuild "QaraamiGen Studio"—a tool for Somali artists to collaborate with AI, preserving the past while composing the future.
Model Architecture Visualization
"The future of tradition is not in repetition, but in evolution."
Listen to Initial Prototypes