Download evaluation/protocol.json from dunli5107/SmollerTTS: direct link, hf CLI and curl.
- Browser
- Download file 1.15 kB
-
https://huggingface.co/dunli5107/SmollerTTS/resolve/main/evaluation/protocol.json
- Command line
-
hf download hf://dunli5107/SmollerTTS/evaluation/protocol.json
-
curl -L -o protocol.json https://huggingface.co/dunli5107/SmollerTTS/resolve/main/evaluation/protocol.json
1.15 kB
| { | |
| "prompt_sha256": "dc165fe8ea76468f4776154a69061be62ddb4bdeeef00c9eeaec41fafe4412a9", | |
| "configurations": [ | |
| "SmollerTTS v28 natural, punctuation/splitting defaults", | |
| "tiny-voca F2 published defaults, seed 0", | |
| "tiny-voca M2 published defaults, seed 0" | |
| ], | |
| "rules": [ | |
| "Use the full text without per-model rewrites.", | |
| "Use every prompt for every configuration; retain failures.", | |
| "Score 16 kHz mono with one pinned UTMOS22 strong checkpoint.", | |
| "Decode all outputs with one pinned Whisper small.en CPU int8 recognizer and the same English normalizer.", | |
| "Report mean UTMOS, corpus WER/CER, generation success and per-category results.", | |
| "Use 10000 stratified prompt-bootstrap replicates, seed 20261119, for 95% intervals and paired differences.", | |
| "Intervals describe this small text panel only, not training-seed or speaker variability.", | |
| "No speed ranking: SmollerTTS uses browser WASM, Tiny Voca native PyTorch.", | |
| "Do not pool this 48-prompt panel with published four-prompt results.", | |
| "No weights, pronunciation rules or defaults may be tuned after seeing these scores." | |
| ] | |
| } | |