Configuration

Configuration

VoxTutor is configured entirely via environment variables. All values have defaults that produce the published benchmark results; no .env file is required to run.

Environment variables

Variable	Default	Description
VOXTUTOR_METHOD	gop	Scorer to use: naive, aligned, normalized, gop, random
VOXTUTOR_REGIME	clean	Distortion regime: clean, warped, noisy
VOXTUTOR_LABELS	(built-in inventory)	Path to a custom phoneme label file (optional)
VOXTUTOR_SAMPLES	40	Number of utterances per cell
VOXTUTOR_SEED	42	Base random seed (harness iterates over 12 seeds from this base)
VOXTUTOR_BACKEND	numpy	Compute backend: numpy (default, offline)

.env.example

# VoxTutor configuration
# Copy to .env and edit as needed. All values are optional (defaults shown).

VOXTUTOR_METHOD=gop
VOXTUTOR_REGIME=clean
# VOXTUTOR_LABELS=       # leave unset for built-in inventory
VOXTUTOR_SAMPLES=40
VOXTUTOR_SEED=42
VOXTUTOR_BACKEND=numpy

Selecting a scorer

# Run one scorer on one regime
VOXTUTOR_METHOD=aligned VOXTUTOR_REGIME=warped voxtutor score

# Or use the CLI flag (overrides env var)
voxtutor score --method aligned --regime warped

Selecting a regime

# List available regimes and their descriptions
voxtutor regimes

# Compare all scorers on a specific regime
voxtutor compare --regime noisy

Reproducibility

Fix VOXTUTOR_SEED and VOXTUTOR_SAMPLES to reproduce exact numbers. The published table uses SEED=42, SAMPLES=40, iterated over 12 seeds (42 through 53). The harness handles this automatically; VOXTUTOR_SEED sets only the single-run seed for voxtutor score and voxtutor compare.