Revolutionizing creative musical expressions using Intel® Tiber™ AI Cloud
In a world where artificial intelligence is revolutionizing industries, an Intel® Student Ambassador has brought AI into the realm of musical creativity. Developed using the Intel® Tiber™ AI Cloud and Intel-optimized software stacks, Chorus.AI is a GenAI solution that composes structured, emotive music from scratch using neural audio synthesis techniques.
The project was tested and accelerated with Intel® Extension for TensorFlow*, enabling significantly faster model inference and seamless music generation on the cloud.
This blog will give you an overview of the Chorus.AI project.
About the Intel® Student Ambassador
Bhumik Kumar Kapoor is currently pursuing a Bachelor of Technology degree at the Bharati Vidyapeeth's College of Engineering in India. As an Intel Student Ambassador, he actively organized impactful workshops on Intel Tiber AI Cloud and Intel® hardware at his university campus. Passionate about AI, creativity, and cloud-native development, they have explored Intel's AI frameworks to create boundary-pushing music generation tools. Chorus.AI is the result of their curiosity and dedication to merging deep learning with sound composition.
Chorus.AI: An Overview
Chorus.AI is a generative AI-based music composer that uses a fine-tuned NSynth model. The model was adapted and deployed using Intel’s AI software stack and Intel Tiber AI Cloud infrastructure. It produces 30-second structured musical compositions with distinct phases like intro, verse, chorus, bridge, and outro. The generator creates original music by processing a sequence of parameters like pitch, velocity, instrument family, and audio features. The model applies phase-wise melodic patterns and stochastic elements like timing jitter and pitch modulation to create expressive musical flows.
Figure 1 and Figure 2 below show how the UI of the project looks like.
Fig.1: UI of Chorus.AI
Fig.2: UI of Chorus.AI
Hardware and Software Specifications
The following configurations were used in the project:
- Cloud Platform: Intel Tiber AI Cloud
- AI Framework: TensorFlow v2.4 with Intel® Extension for TensorFlow
- AI Model: based on Bi-LSTM
- Hardware: Intel® Xeon® Scalable processors
- Programming Language: Python* v3.9
- Others: pretty_midi, NumPy*, Scikit-learn*, Streamlit app and Three.js library (for UI development)
System Architecture and Workflow
Figure 3 below shows the building blocks of Chorus.AI. It uses the NSynth model developed by the Google* Magenta team.
Fig.3: Block diagram of Chorus.AI’s NSynth-based melody studio
The tool follows the following steps for music composition:
- Seed Initialization: Musical seeds are created with normalized pitch, velocity, and instrument metadata.
- LSTM-Based Generation: A trained NSynth LSTM generates future sequences with optional Gaussian noise for variety.
- Phase Composition: The song is structured into multiple emotional and melodic phases.
- MIDI Rendering: Notes are translated into MIDI format using pretty_midi with precise velocity and timing.
Performance Optimization Using Intel® AI Software
By using Intel Extension for TensorFlow on the Tiber AI Cloud, a significant reduction in model inference time was achieved.
- Original inference time: ~3-4 hrs
- Optimized inference time: < 1 hr
The >90% reduction in inference time enabled rapid experimentation and real-time testing of musical ideas.[1]
Sample Output
A 30-second, six-phase AI-generated song is successfully produced, each phase featuring unique instruments and melodies:
|
The composition is exported as melodic_30sec_song_with_variations.mid.
RESULTS
Fig.4: MAE (Mean Absolute Error) and MSE (Mean Squared Error) results visualization
The Huber loss function combines the best properties of both MAE and MSE.
Fig.5: Huber loss visualization
Fig.6: Note density visualization
Fig.7: Pitch class distribution
Potential Real-world Applications of Chorus.AI
Chorus.AI's flexible architecture allows its use across various industries. Example use cases of the project include:
- Film and TV scoring
- Interactive gaming soundtracks
- Therapeutic music generation
- Educational tools for music theory and AI
Future Roadmap of the Project
The student ambassador plans to enhance the project by:
- Deploying it on Intel® Gaudi® 2 AI accelerators,
- Fine-tuning with alternate datasets like NES-MDB, and
- Emotion-driven composition using real-time mood classifiers.
|
Conclusion
Chorus.AI showcases the synergy between AI and musical artistry. It leverages Intel’s cloud platform and software ecosystem to redefine how music can be created. From idea to implementation, this project is a demonstration of how accessible and powerful GenAI becomes with Intel Tiber AI Cloud.
Useful Resources
- Intel Tiber AI Cloud
- TensorFlow Optimizations from Intel
- GenAI Developer Resources
- Jump-start AI Development
- Intel Student Ambassador Program
Disclaimer
Intel does not control or audit third-party data. You should consult other sources to evaluate accuracy.
Intel technologies may require enabled hardware, software, or service activation.
Your costs and results may vary.
Intel does not control or audit third-party data. You should consult other sources to evaluate accuracy.
[1] Performance results are based on testing done using the specified configurations as of Mar-Apr 2025 and may not reflect all publicly available updates. See backup for configuration details. No product or component can be absolutely secure.
Technical Software Product Marketing Engineer, Intel