Fix matcha tts zh-en model (k2-fsa#2851)

Author: csukuangfj

scripts/matcha-tts/zh-en/generate_lexicon.py

@@ -1,6 +1,6 @@
 #!/usr/bin/env python3
 
-from pypinyin import Style, lazy_pinyin, load_phrases_dict, phrases_dict, pinyin_dict
+from pypinyin import Style, pinyin, load_phrases_dict, phrases_dict, pinyin_dict
 
 load_phrases_dict(
     {
@@ -28,7 +28,8 @@ def main():
                 continue
 
             w = chr(key)
-            tokens = lazy_pinyin(w, style=Style.TONE3, tone_sandhi=True)[0]
+            tokens = pinyin(w, style=Style.TONE3, neutral_tone_with_five=True)[0][0]
+
             if tokens == "shei2":
                 tokens = "shui2"
 
@@ -43,15 +44,17 @@ def main():
         for key in phrases:
             if key in user_defined:
                 continue
-            tokens = lazy_pinyin(key, style=Style.TONE3, tone_sandhi=True)
+            tokens = pinyin(key, style=Style.TONE3, neutral_tone_with_five=True)
             for i in range(len(tokens)):
                 if tokens[i] == "shei2":
                     tokens[i] = "shui2"
 
                 if tokens[i][-1] not in ("1", "2", "3", "4", "5"):
                     tokens[i] += "1"
 
-            tokens = " ".join(tokens)
+            flatten = [t[0] for t in tokens]
+
+            tokens = " ".join(flatten)
 
             f.write(f"{key} {tokens}\n")
 

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

@@ -266,10 +266,6 @@ class MatchaTtsLexicon::Impl {
       }
     }
 
-    if (IsAlphaOrPunct(w.front())) {
-      ans.push_back(token2id_.at(" "));
-    }
-
     if (debug_) {
       std::ostringstream os;
       os << w << ": ";

sherpa-onnx/csrc/offline-tts-matcha-impl.h

@@ -431,8 +431,16 @@ class OfflineTtsMatchaImpl : public OfflineTtsImpl {
 
     std::vector<int64_t> x;
     x.reserve(num_tokens);
-    for (const auto &k : tokens) {
-      x.insert(x.end(), k.begin(), k.end());
+    if (config_.model.debug) {
+      for (const auto &k : tokens) {
+        x.insert(x.end(), k.begin(), k.end());
+      }
+      std::ostringstream oss;
+      for (int32_t i : x) {
+        oss << i << ", ";
+      }
+      oss << "\n";
+      SHERPA_ONNX_LOGE("%s\n", oss.str().c_str());
     }
 
     auto memory_info =