Understanding Patch Generation
The pipeline
Section titled “The pipeline”PROMPT
Natural-language description, submitted from the app
SEMANTIC PARSING
Text is mapped onto acoustic-feature vectors
PARAMETER MATRIX
The cognitive mapper distributes weights across the engine layer
VALIDATION
Result is clamped to the real parameter schema before it reaches you
PATCH
Returned as {patch, validation, slug}
Where this runs
Section titled “Where this runs”Generation is a cloud round-trip, not a local model: your prompt is sent to
POST /api/v1/murmur/generate-preset on the Murmur API (Go), which hands
semantic parsing and parameter-matrix work to the Python intelligence
worker. See Murmur Cloud for the service split and
API reference for the endpoint
contract.
Why generation can miss
Section titled “Why generation can miss”The mapper is translating language into a fixed, finite parameter space — it can’t invent a synthesis technique the engine layer doesn’t have. If a generated patch doesn’t match what you described, see Troubleshooting Patch Generation for how to get a closer result.