diff --git a/sherpa-onnx/csrc/online-websocket-server-impl.cc b/sherpa-onnx/csrc/online-websocket-server-impl.cc
index a47e613c86..ca9f2bf854 100644
--- a/sherpa-onnx/csrc/online-websocket-server-impl.cc
+++ b/sherpa-onnx/csrc/online-websocket-server-impl.cc
@@ -19,12 +19,17 @@ void OnlineWebsocketDecoderConfig::Register(ParseOptions *po) {
 
   po->Register("max-batch-size", &max_batch_size,
                "Max batch size for recognition.");
+
+  po->Register("end-tail-padding", &end_tail_padding,
+               "It determines the length of tail_padding at the end of audio.");
 }
 
 void OnlineWebsocketDecoderConfig::Validate() const {
   recognizer_config.Validate();
   SHERPA_ONNX_CHECK_GT(loop_interval_ms, 0);
   SHERPA_ONNX_CHECK_GT(max_batch_size, 0);
+  SHERPA_ONNX_CHECK_GT(end_tail_padding, 0);
+
 }
 
 void OnlineWebsocketServerConfig::Register(sherpa_onnx::ParseOptions *po) {
@@ -82,8 +87,7 @@ void OnlineWebsocketDecoder::InputFinished(std::shared_ptr<Connection> c) {
     c->samples.pop_front();
   }
 
-  // TODO(fangjun): Change the amount of paddings to be configurable
-  std::vector<float> tail_padding(static_cast<int64_t>(0.8 * sample_rate));
+  std::vector<float> tail_padding(static_cast<int64_t>(config_.end_tail_padding * sample_rate));
 
   c->s->AcceptWaveform(sample_rate, tail_padding.data(), tail_padding.size());
 
diff --git a/sherpa-onnx/csrc/online-websocket-server-impl.h b/sherpa-onnx/csrc/online-websocket-server-impl.h
index a82170fbaa..9716c5c722 100644
--- a/sherpa-onnx/csrc/online-websocket-server-impl.h
+++ b/sherpa-onnx/csrc/online-websocket-server-impl.h
@@ -62,6 +62,8 @@ struct OnlineWebsocketDecoderConfig {
 
   int32_t max_batch_size = 5;
 
+  float end_tail_padding = 0.8;
+
   void Register(ParseOptions *po);
   void Validate() const;
 };