Engineering Docs

Study how neural audio systems represent sound, learn conditional relationships, generate audio, and are evaluated in production and research settings.

Follow the path

Work through the material in three layers. The sequence moves from signal foundations to model design and then to production systems.

1. Represent audio

1. Digital Audio Basics — establish the signal-level vocabulary used throughout the documentation.
2. Music Representations — compare waveforms, spectrograms, symbolic formats, embeddings, and codec tokens.
3. Signal Processing Basics — connect time-domain signals to frequency-domain analysis.

2. Model and train

4. Transformers for Audio and Diffusion Models — examine core generation architectures.
5. Dataset Curation — understand the data, rights, and labeling decisions that shape a model.
6. Evaluation Metrics — combine computational measurements with structured listening tests.

3. Build a system

Apply the foundations through APIs and Integration, AI Music Agents, and Advanced Topics. Use Reliability and Sourcing when documenting claims and system behavior.

Choose by engineering task

I need to…	Start here
Design an audio representation	Audio Embeddings and Neural Audio Codecs
Select a model architecture	Architecture
Prepare or augment training data	Training
Integrate a hosted model	APIs
Build a multi-model system	AI Music Agents
Fine-tune or control a model	Advanced Topics
Check terminology	Glossary

These pages assume comfort with software concepts and introduce mathematics where it clarifies system behavior. Readers primarily interested in making music can use the User Guides.