File formats

Presets

Plain text, one key = value per line; see Presets.

Saved memory

Save Memory… (and mse-render --save-memory) writes a folder:

My memory/
├── manifest.json
├── trace_0001.wav
├── trace_0002.wav
└── …

Each trace_NNNN.wav is one trace’s audio, 32-bit float, at the session’s sample rate, oldest first. manifest.json describes the memory:

{
  "format": "MINERVA Space Echo memory",
  "version": 1,
  "sample_rate": 48000,
  "channels": 2,
  "capacity": 100,
  "feature_slots": 16,
  "feature_bands": 24,
  "params": { "power": 3, "capacity": 100, "...": "..." },
  "traces": [
    {
      "file": "trace_0001.wav",
      "begin": 0,
      "length": 96000,
      "nominal_length": 96000,
      "serial": 1,
      "rms": 0.12,
      "strength": 1.0,
      "use_count": 2.4,
      "generation": 0,
      "merge_count": 1,
      "clamped": false,
      "features": [0.41, -1.2, "… 384 values …"],
      "context": [0.12, 0.9, "… 384 values (optional) …"]
    }
  ]
}

Field	Meaning
`begin`	Where the audio starts within its segment, in samples (a trace recorded from mid-segment starts later).
`length`	Samples of audio in the WAV.
`nominal_length`	The trace length setting when it was recorded, in samples (used by Length Mismatch).
`serial`	Order of storage; ages are counted from it.
`rms`	The trace’s level.
`strength`	Activation multiplier; fades with Fading / Segment.
`use_count`	Recent accumulated activation (for When Full = Least Used).
`generation`	0 = recorded from the input; n = an echo of generation n−1 material.
`merge_count`	How many segments were consolidated into it.
`clamped`	Locked against replacement.
`features`	The address: 16 slots × 24 bands, slot-major (`index = slot × 24 + band`).
`context`	Optional: the previous segment’s address (the n−1 half, same layout). Missing means no context.
`params`	The parameter values when the memory was saved (for reference; loading memory doesn’t change settings).

Loading resamples memory saved at another sample rate. Newer format versions are refused by older plug-ins. To add new audio to memory, use Import Audio… rather than editing the folder: imported traces get correct addresses.