diff --git a/conformance/tha-phonemic-vectors.json b/conformance/tha-phonemic-vectors.json new file mode 100644 index 00000000..56aad418 --- /dev/null +++ b/conformance/tha-phonemic-vectors.json @@ -0,0 +1,58 @@ +{ + "model": "tha-phonemic-1.0", + "index_version": 12, + "decode": { + "mode": "single-row-greedy" + }, + "source": "run-080-thai-lex-v6a oepoch-3 (clean-split PER 19.29)", + "vectors": [ + { + "src": "ก", + "out": "k ɔː ˧" + }, + { + "src": "ก กา", + "out": "k a ˨˩ . k aː ˧" + }, + { + "src": "ก ข ไม่กระดิกหู", + "out": "k o k̚ ˨˩ . kʰ a ˨˩ . k a ˧ . k uː ˧" + }, + { + "src": "ก หัน", + "out": "k a ˨˩ . h a n ˩˩˦" + }, + { + "src": "กก", + "out": "k o k̚ ˨˩" + }, + { + "src": "กกช้าง", + "out": "k o k̚ ˨˩ . t͡ɕʰ aː ŋ ˦˥" + }, + { + "src": "กกธูป", + "out": "k o k̚ ˨˩ . kʰ uː p̚ ˥˩" + }, + { + "src": "กกหู", + "out": "k o k̚ ˨˩ . h uː ˩˩˦" + }, + { + "src": "กกุธภัณฑ์", + "out": "k a ˨˩ . k u t̚ ˨˩ . pʰ a n ˧" + }, + { + "src": "กกเสา", + "out": "k o k̚ ˨˩ . s a w ˩˩˦" + }, + { + "src": "กง", + "out": "k o ŋ ˧" + }, + { + "src": "กงกอน", + "out": "k o ŋ ˧ . k ɔː n ˧" + } + ] +} \ No newline at end of file diff --git a/docs/catalogue/models-catalogue.json b/docs/catalogue/models-catalogue.json index 657d607a..20154fcb 100644 --- a/docs/catalogue/models-catalogue.json +++ b/docs/catalogue/models-catalogue.json @@ -722,6 +722,23 @@ ], "status": "shipped" }, + { + "id": "tha-phonemic-1.0", + "task": "g2p", + "precision": "fp32", + "license": "BSD-3-Clause", + "size_bytes": null, + "sha256": null, + "metrics": [ + { + "name": "per", + "value": 19.29, + "protocol": "phoneme-token PER, clean v1 test split (3,210 rows), single-row greedy", + "anchor": null + } + ], + "status": "release-candidate" + }, { "id": "urd-diac-1.0", "task": "diacritization", diff --git a/models.yaml b/models.yaml index d1a7d18d..8b397703 100644 --- a/models.yaml +++ b/models.yaml @@ -1,6 +1,6 @@ version: 1 models: - tha-g2p-lex-1.0: + tha-phonemic-1.0: task: g2p precision: fp32 url: https://github.com/interscript/interscript-models/releases/download/UNPUBLISHED/tha-g2p-lex-1.0.zip