AI-Powered Song Recognition and Music Generation
The project aimed to develop an AI-driven audio platform featuring two primary functions: a real-time music recognition system and a songwriting assistant that can generate original music on demand. The recognition system quickly identifies songs, akin to Shazam, while the AI music generation feature allows users to request complete songs composed by AI. The input consists of audio (recorded or live), and the outputs are the recognized song and an original track generated by the AI.
**Technologies & Skills:**
- Artificial Intelligence
- Generative AI
- Deep Learning
- Audio AI
- Signal Processing
- Programming Language: Python
**Project Development Approach:**
For the song recognition component, the team utilized two methodologies: traditional signal processing and deep learning. Audio fingerprints were derived from spectrograms, and a Siamese neural network with a triplet loss function was used to develop robust embeddings (128-dim) for effective matching. Various classifiers were then employed to ensure accurate song identification. In terms of song generation, the team's emphasis was on creating a cost-effective generative model that achieves quality comparable to Suno without incurring high cloud expenses. This model was trained on a broad and varied dataset to generate full-length songs (up to 4 minutes), complete with both vocals and musical accompaniment. The architecture is designed for scalability, efficiency, and adaptability for future feature enhancements, ensuring its sustainability.
**Solution & Results:**
The final solution includes a real-time music recognition engine that can swiftly identify tracks using a moderate dataset. Concurrently, an AI-based song generation prototype successfully creates original tracks featuring vocals and instrumental backing. This solution opens new avenues for innovation in the music technology sector, applicable in consumer applications, professional creative tools, music streaming services, entertainment apps, social media, and educational platforms.