AMLP categorizes nodes into six fundamental types, each with distinct characteristics and runtime requirements.
DSP Nodes
Traditional Signal Processing
—Deterministic computation — same input always produces same output
—Low latency — typically <1ms, often zero-latency for IIR filters
—Minimal memory — state size proportional to filter order or delay length
—CPU-only processing — usually sufficient, GPU acceleration rarely needed
—Mathematical provability — frequency response and phase characteristics computable analytically
Examples: dsp.biquad, dsp.compressor, dsp.beamformer, dsp.fir, dsp.reverb
ML Nodes
Neural Network Inference
—Non-deterministic training — model behavior learned from data
—Higher latency — typically 5-50ms depending on model size and hardware
—Significant memory — model weights, activation buffers, quantization tables
—Hardware acceleration — often benefit from GPU/NPU/DSP acceleration
—Quality-compute tradeoffs — larger models generally better but slower
Examples: ml.model, ml.ensemble, ml.conditional, ml.transformer, ml.conformer
Hybrid Nodes
DSP-ML Integration
—Best of both worlds — DSP reliability with ML adaptability
—Lower latency — DSP in critical path, ML for parameter adaptation
—Interpretable parameters — DSP parameters have physical meaning
—Graceful degradation — falls back to DSP if ML fails
Examples: hybrid.adaptive_filter, hybrid.neural_effect, hybrid.ml_compressor
Spatial Audio Nodes
Immersive Audio Processing
—Object-based processing — audio objects with spatial metadata (position, size, divergence)
—Scene-based encoding — ambisonics (First Order, Higher Order)
—Binaural rendering — HRTF convolution for headphone reproduction
—Format conversion — translation between spatial audio formats
—Downmix strategies — adaptive conversion to various speaker configurations
Examples: spatial.object_renderer, spatial.ambisonics_encoder, spatial.binaural_renderer, spatial.iamf_encoder
Network Audio Nodes
Audio-Over-IP Integration
—AES67 compliance — open standard for audio-over-IP interoperability
—SMPTE ST 2110-30 — broadcast standard for professional media over IP
—PTP synchronization — IEEE 1588 sample-accurate timing
—QoS management — network quality of service for guaranteed delivery
—Distributed processing — ML inference across networked systems
Examples: network.aes67_input, network.aes67_output, network.st2110_interface, network.ptp_sync
MCP Integration Nodes
AI Ecosystem Connectivity
—Resource exposure — expose audio models and pipelines as MCP resources
—Tool invocation — allow AI systems to trigger audio processing via MCP tools
—Context provision — provide audio analysis results to AI assistants
—Bidirectional communication — enable AI-driven parameter control
Examples: mcp.resource_server, mcp.tool_provider, mcp.context_bridge, mcp.ai_assistant_interface