Card: English detection convention and switch

README.md

@@ -46,7 +46,7 @@ Defaults: **temperature 1.0**, **tempo 0.8** (`length_scale`; at 1.0 v2 speaks a
 | `onnx/voices.json` | the male_voxcpm_1 speaker embedding, 18 style embeddings, defaults |
 | `model/matcha_pl_v2.ckpt` | PyTorch checkpoint (Matcha-TTS + the tts-pl patch: style tokens, Polish cleaners) — speaker 20 = male_voxcpm_1 |
 | `vocoder/generator_male_voxcpm_1`, `vocoder/config.json` | HiFi-GAN (v1 architecture) fine-tuned on this model's mels ↔ the male_voxcpm_1 audio |
-| `english/` | English detection (`english_detect.py`, `english_words.txt`; browser: `english.js`, `english_data.json`) and the mixed phonemizer (`polish_mixed_cleaners.py`; browser: `cleaner.js`) |
+| `english/` | English detection (`english_detect.py`, `english_words.txt`, `english_phrase_words.txt`; browser: `english.js`, `english_data.json`) and the mixed phonemizer (`polish_mixed_cleaners.py`; browser: `cleaner.js`). A word that is also a Polish word (SGJP dictionary via [Morfeusz 2](http://morfeusz.sgjp.pl/), BSD-2; e.g. *sale*, *to go*) is read as Polish unless it is listed on its own in `english_words.txt` or written in braces: `{sale}`. Detection can be switched off: everything is then read as Polish, braces still mark English. |
 | `samples/` | the 12 held-out sentences, v2 and teacher |
 | `data/` | style token map, speaker map |
 | `ATTRIBUTION.md` | every text source of the training data, and the base model's data |