Optimize Speaker EQ (Spinorama)

Every speaker has frequency response imperfections. SotF can automatically correct these using real measurement data from spinorama.org, a database of 800+ speakers measured under controlled conditions (CEA-2034 / Spinorama standard).

What you need

SotF (Terminal or Desktop)
Your speaker model name (must be in the spinorama.org database)
Internet connection (for fetching measurement data)

Open the Spinorama EQ screen

Click the Spinorama tab in the sidebar.
Search for your speaker

Type your speaker model (e.g., “KEF R3”, “JBL 305P”, “Genelec 8030”). Select from the search results. SotF fetches the measurement data from spinorama.org.
Select measurement version

Some speakers have multiple measurements (e.g., from different reviewers like ASR, EAC, Princeton). Choose the one that matches your speaker revision.

Configure optimization (optional)

Parameter	Default	Description
Number of filters	7	PEQ filter count
Loss function	speaker-score	Optimizes Harman preference score
Algorithm	cobyla	Local optimizer (fast). Use `autoeq:de` for global search
Min/Max Q	0.5 / 10.0	Filter sharpness bounds
Min spacing (oct)	0.3	Minimum distance between filters

Run optimization

Click Optimize. Watch the loss curve converge. This typically takes 5-30 seconds depending on algorithm choice.
Review and apply

Inspect the before/after frequency response overlay. Click Apply to load the correction EQ into your plugin chain.

# Optimize for KEF R3 using COBYLA
cargo run --bin autoeq --release -- \
  --speaker "KEF R3" \
  --version asr \
  --measurement CEA2034 \
  --loss speaker-score \
  --algo nlopt:cobyla \
  -n 7

# Global search with Differential Evolution (slower, better for complex responses)
cargo run --bin autoeq --release -- \
  --speaker "KEF R3" \
  --version asr \
  --measurement CEA2034 \
  --loss speaker-score \
  --algo autoeq:de \
  --population 50 \
  --iterations 5000 \
  -n 10

Loss functions explained

Loss	Best for	What it optimizes
`speaker-flat`	Studio monitors	Minimizes deviation from perfectly flat response
`speaker-score`	Listening preference	Optimizes the Harman/Olive preference score (slight bass boost + PIR flatness)

Algorithm choice

Algorithm	Speed	Quality	Best for
`nlopt:cobyla`	Fast (seconds)	Good	Quick optimization, simple responses
`autoeq:de`	Slow (minutes)	Excellent	Complex responses, many filters
`ga`	Medium	Good	Alternative global search
`pso`	Medium	Good	Particle swarm, good for exploration