---

name: data-and-config description: "Prepare and validate StyleTTS2 data lists, OOD text files, config YAML, and asset paths before training or inference." disable-model-invocation: true metadata: disco-role: operating license: MIT

data-and-config

Use this sub-skill to check the inputs StyleTTS2 workflows depend on before any stage launch or demo handoff.

Covers

• train/val list schema, speaker labels, and OOD text files
• 24 kHz audio-root assumptions and validation split handling
• config fields that drive stage selection, checkpoint loading, model layout, and memory use
• safe inspectors:
  • scripts/validate_data_lists.py checks list rows, optional wav existence, speaker ids, and OOD text readiness without training.
  • scripts/inspect_config.py summarizes high-impact config fields, path semantics, and stage warnings without launching a run.

Route elsewhere

• Training launches, checkpointed stage progression, and resume behavior: training sub-skill
• Pretrained demo assets, reference-audio handling, and synthesis setup: inference sub-skill

Quick checks

• Validate rows and optional file presence before training.
• Inspect a config before editing paths, batch size, decoder type, or checkpoint settings.
• Keep explicit speaker ids in public lists, especially for multispeaker data.

References

• Data formats
• Configuration guide
• Troubleshooting