Preserve previous result as context for next segment (#1335)

Co-authored-by: vsd-vector <askars.salimbajevs@tilde.lv>

@@ -360,11 +360,13 @@ class OnlineRecognizerTransducerImpl : public OnlineRecognizerImpl {
   }
 
   void Reset(OnlineStream *s) const override {
+    int32_t context_size = model_->ContextSize();
+
     {
       // segment is incremented only when the last
-      // result is not empty
+      // result is not empty, contains non-blanks and longer than context_size)
       const auto &r = s->GetResult();
-      if (!r.tokens.empty() && r.tokens.back() != 0) {
+      if (!r.tokens.empty() && r.tokens.back() != 0 && r.tokens.size() > context_size) {
         s->GetCurrentSegment() += 1;
       }
     }
@@ -372,10 +374,6 @@ class OnlineRecognizerTransducerImpl : public OnlineRecognizerImpl {
     // reset encoder states
     // s->SetStates(model_->GetEncoderInitStates());
 
-    // we keep the decoder_out
-    decoder_->UpdateDecoderOut(&s->GetResult());
-    Ort::Value decoder_out = std::move(s->GetResult().decoder_out);
-
     auto r = decoder_->GetEmptyResult();
     if (config_.decoding_method == "modified_beam_search" &&
         nullptr != s->GetContextGraph()) {
@@ -383,8 +381,19 @@ class OnlineRecognizerTransducerImpl : public OnlineRecognizerImpl {
         it->second.context_state = s->GetContextGraph()->Root();
       }
     }
+
+    auto last_result = s->GetResult();
+    // if last result is not empty, then
+    // preserve last tokens as the context for next result
+    if (static_cast<int32_t>(last_result.tokens.size()) > context_size) {
+      std::vector<int64_t> context(last_result.tokens.end() - context_size, last_result.tokens.end());
+
+      Hypotheses context_hyp({{context, 0}});
+      r.hyps = std::move(context_hyp);
+      r.tokens = std::move(context);
+    }
+
     s->SetResult(r);
-    s->GetResult().decoder_out = std::move(decoder_out);
 
     // Note: We only update counters. The underlying audio samples
     // are not discarded.