Use piper-phonemize to convert text to token IDs (#453)

sherpa-onnx/csrc/offline-tts.h

@@ -28,10 +28,17 @@ struct OfflineTtsConfig {
   // If there are multiple rules, they are applied from left to right.
   std::string rule_fsts;
 
+  // Maximum number of sentences that we process at a time.
+  // This is to avoid OOM for very long input text.
+  // If you set it to -1, then we process all sentences in a single batch.
+  int32_t max_num_sentences = 2;
+
   OfflineTtsConfig() = default;
   OfflineTtsConfig(const OfflineTtsModelConfig &model,
-                   const std::string &rule_fsts)
-      : model(model), rule_fsts(rule_fsts) {}
+                   const std::string &rule_fsts, int32_t max_num_sentences)
+      : model(model),
+        rule_fsts(rule_fsts),
+        max_num_sentences(max_num_sentences) {}
 
   void Register(ParseOptions *po);
   bool Validate() const;