We default to the 6-layer model for every language that has one (en/it/de/es/pt); French is forced to 24-layer since pocket-tts ships no 6-layer French. Some languages also have higher-quality 24-layer (_24l) variants (~3× larger, ~4× slower) we don't currently expose.
Question: do we need a per-language quality toggle (POCKET_MODEL_QUALITY=fast|high + per-language override), or is 6-layer-default good enough? Parked design exists (settings + resolver + configure.sh step + graceful fallback for en/fr single-variant cases).
Decide based on whether anyone wants the higher-quality 24-layer output for es/de/it/pt.
We default to the 6-layer model for every language that has one (
en/it/de/es/pt); French is forced to24-layersincepocket-ttsships no6-layerFrench. Some languages also have higher-quality 24-layer (_24l) variants (~3× larger,~4× slower) we don't currently expose.Question: do we need a per-language quality toggle (
POCKET_MODEL_QUALITY=fast|high+ per-language override), or is 6-layer-default good enough? Parked design exists (settings + resolver +configure.shstep + graceful fallback foren/frsingle-variant cases).Decide based on whether anyone wants the higher-quality 24-layer output for
es/de/it/pt.