EngineX-Iluvatar/enginex-mr_series-sherpa-onnx
Archived
16
0
Fork 0
Code Issues Pull Requests Actions Projects Releases Wiki Activity

Fix looking up OOVs in lexicon.txt for MeloTTS models. (#1266)

Browse Source 
If an English word does not exist in the lexicon, we split
it into characters. For instance, if the word TTS does not
exist in lexicon.txt, we split it into 3 characters T, T, and S.
This commit is contained in:
Fangjun Kuang
2024-08-16 22:10:03 +08:00
committed by GitHub
parent 63713ecbf0
commit 9dcea49dba
1 changed files with 16 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
sherpa-onnx/csrc/melo-tts-lexicon.cc
View File

@@ -136,6 +136,22 @@ class MeloTtsLexicon::Impl {
ans.tokens.insert(ans.tokens.end(), ids.tokens.begin(), ans.tokens.insert(ans.tokens.end(), ids.tokens.begin(),
ids.tokens.end()); ids.tokens.end());
ans.tones.insert(ans.tones.end(), ids.tones.begin(), ids.tones.end()); ans.tones.insert(ans.tones.end(), ids.tones.begin(), ids.tones.end());
} else {
// If the lexicon does not contain the word, we split the word into
// characters.
//
// For instance, if the word is TTS and it is does not exist
// in the lexicon, we split it into 3 characters: T T S
std::string s;
for (char c : word) {
s = c;
if (word2ids_.count(s)) {
const auto &t = word2ids_.at(s);
ans.tokens.insert(ans.tokens.end(), t.tokens.begin(),
t.tokens.end());
ans.tones.insert(ans.tones.end(), t.tones.begin(), t.tones.end());
}
}
} }
} }