Fix(csrc/melotts): Fix V-words pronounciation on MeloTTS_en (#3002)

Author: pqsworld

scripts/melo-tts/test.py

@@ -15,6 +15,11 @@ def __init__(self, lexion_filename: str, tokens_filename: str):
             for line in f:
                 s, i = line.split()
                 tokens[s] = int(i)
+        # Map "v" to "V" token ID (same as post_replace_ph in MeloTTS, only for English models)
+        # English models have "V" with token ID 14
+        if tokens.get("V") == 14 and "v" in tokens:
+            tokens["v"] = tokens["V"]
+
 
         lexicon = dict()
         with open(lexion_filename, encoding="utf-8") as f:

sherpa-onnx/csrc/melo-tts-lexicon.cc

@@ -228,6 +228,12 @@ class MeloTtsLexicon::Impl {
     if (!token2id_.count("、") && token2id_.count("，")) {
       token2id_["、"] = token2id_["，"];
     }
+
+    // Map 'v' to 'V' token (same as post_replace_ph in MeloTTS)
+    // Only for English models
+    if (meta_data_.language == "en" && token2id_.count("V")) {
+      token2id_["v"] = token2id_["V"];
+    }
   }
 
   void InitLexicon(std::istream &is) {