diff --git a/cmake/onnxruntime-linux-riscv64-static.cmake b/cmake/onnxruntime-linux-riscv64-static.cmake
index 8e7c2611f..b400c4741 100644
--- a/cmake/onnxruntime-linux-riscv64-static.cmake
+++ b/cmake/onnxruntime-linux-riscv64-static.cmake
@@ -16,7 +16,7 @@ endif()
 
 set(onnxruntime_URL  "https://github.com/csukuangfj/onnxruntime-libs/releases/download/v1.18.0/onnxruntime-linux-riscv64-static_lib-1.18.0.zip")
 set(onnxruntime_URL2 "https://hub.nuaa.cf/csukuangfj/onnxruntime-libs/releases/download/v1.18.0/onnxruntime-linux-riscv64-static_lib-1.18.0.zip")
-set(onnxruntime_HASH "SHA256=6791d695d17118db6815364c975a9d7ea9a8909754516ed1b089fe015c20912e")
+set(onnxruntime_HASH "SHA256=77ecc51d8caf0953755db6edcdec2fc03bce3f6d379bedd635be50bb95f88da5")
 
 # If you don't have access to the Internet,
 # please download onnxruntime to one of the following locations.
diff --git a/cmake/onnxruntime-linux-riscv64.cmake b/cmake/onnxruntime-linux-riscv64.cmake
index 748a2bb52..f84b63687 100644
--- a/cmake/onnxruntime-linux-riscv64.cmake
+++ b/cmake/onnxruntime-linux-riscv64.cmake
@@ -16,7 +16,7 @@ endif()
 
 set(onnxruntime_URL  "https://github.com/csukuangfj/onnxruntime-libs/releases/download/v1.18.0/onnxruntime-linux-riscv64-1.18.0.zip")
 set(onnxruntime_URL2 "https://hub.nuaa.cf/csukuangfj/onnxruntime-libs/releases/download/v1.18.0/onnxruntime-linux-riscv64-1.18.0.zip")
-set(onnxruntime_HASH "SHA256=87ef36dbba28ee332069e7e511dcb409913bdeeed231b45172fe200d71c690a2")
+set(onnxruntime_HASH "SHA256=81a11b54d1d71f4b3161b00cba8576a07594abd218aa5c0d82382960ada06092")
 
 # If you don't have access to the Internet,
 # please download onnxruntime to one of the following locations.
diff --git a/sherpa-onnx/csrc/provider.cc b/sherpa-onnx/csrc/provider.cc
index 9c50eb8cc..dcf3a261d 100644
--- a/sherpa-onnx/csrc/provider.cc
+++ b/sherpa-onnx/csrc/provider.cc
@@ -20,6 +20,8 @@ Provider StringToProvider(std::string s) {
     return Provider::kCUDA;
   } else if (s == "coreml") {
     return Provider::kCoreML;
+  } else if (s == "xnnpack") {
+    return Provider::kXnnpack;
   } else {
     SHERPA_ONNX_LOGE("Unsupported string: %s. Fallback to cpu", s.c_str());
     return Provider::kCPU;
diff --git a/sherpa-onnx/csrc/provider.h b/sherpa-onnx/csrc/provider.h
index 8e0dcc0a1..05ce71654 100644
--- a/sherpa-onnx/csrc/provider.h
+++ b/sherpa-onnx/csrc/provider.h
@@ -13,9 +13,10 @@ namespace sherpa_onnx {
 // https://github.com/microsoft/onnxruntime/blob/main/java/src/main/java/ai/onnxruntime/OrtProvider.java
 // for a list of available providers
 enum class Provider {
-  kCPU = 0,     // CPUExecutionProvider
-  kCUDA = 1,    // CUDAExecutionProvider
-  kCoreML = 2,  // CoreMLExecutionProvider
+  kCPU = 0,      // CPUExecutionProvider
+  kCUDA = 1,     // CUDAExecutionProvider
+  kCoreML = 2,   // CoreMLExecutionProvider
+  kXnnpack = 3,  // XnnpackExecutionProvider
 };
 
 /**
diff --git a/sherpa-onnx/csrc/session.cc b/sherpa-onnx/csrc/session.cc
index 6080bae27..827c4e05b 100644
--- a/sherpa-onnx/csrc/session.cc
+++ b/sherpa-onnx/csrc/session.cc
@@ -25,6 +25,12 @@ static Ort::SessionOptions GetSessionOptionsImpl(int32_t num_threads,
   sess_opts.SetIntraOpNumThreads(num_threads);
   sess_opts.SetInterOpNumThreads(num_threads);
 
+  std::vector<std::string> available_providers = Ort::GetAvailableProviders();
+  std::ostringstream os;
+  for (const auto &ep : available_providers) {
+    os << ep << ", ";
+  }
+
   // Other possible options
   // sess_opts.SetGraphOptimizationLevel(ORT_ENABLE_EXTENDED);
   // sess_opts.SetLogSeverityLevel(ORT_LOGGING_LEVEL_VERBOSE);
@@ -33,9 +39,17 @@ static Ort::SessionOptions GetSessionOptionsImpl(int32_t num_threads,
   switch (p) {
     case Provider::kCPU:
       break;  // nothing to do for the CPU provider
+    case Provider::kXnnpack: {
+      if (std::find(available_providers.begin(), available_providers.end(),
+                    "XnnpackExecutionProvider") != available_providers.end()) {
+        sess_opts.AppendExecutionProvider("XNNPACK");
+      } else {
+        SHERPA_ONNX_LOGE("Available providers: %s. Fallback to cpu!",
+                         os.str().c_str());
+      }
+      break;
+    }
     case Provider::kCUDA: {
-      std::vector<std::string> available_providers =
-          Ort::GetAvailableProviders();
       if (std::find(available_providers.begin(), available_providers.end(),
                     "CUDAExecutionProvider") != available_providers.end()) {
         // The CUDA provider is available, proceed with setting the options
@@ -47,8 +61,9 @@ static Ort::SessionOptions GetSessionOptionsImpl(int32_t num_threads,
         sess_opts.AppendExecutionProvider_CUDA(options);
       } else {
         SHERPA_ONNX_LOGE(
-            "Please compile with -DSHERPA_ONNX_ENABLE_GPU=ON. Fallback to "
-            "cpu!");
+            "Please compile with -DSHERPA_ONNX_ENABLE_GPU=ON. Available "
+            "providers: %s. Fallback to cpu!",
+            os.str().c_str());
       }
       break;
     }
diff --git a/sherpa-onnx/csrc/sherpa-onnx-vad-microphone.cc b/sherpa-onnx/csrc/sherpa-onnx-vad-microphone.cc
index 01f8d7e5e..19dd1d85f 100644
--- a/sherpa-onnx/csrc/sherpa-onnx-vad-microphone.cc
+++ b/sherpa-onnx/csrc/sherpa-onnx-vad-microphone.cc
@@ -43,8 +43,8 @@ This program shows how to use VAD in sherpa-onnx.
 
   ./bin/sherpa-onnx-vad-microphone \
     --silero-vad-model=/path/to/silero_vad.onnx \
-    --provider=cpu \
-    --num-threads=1
+    --vad-provider=cpu \
+    --vad-num-threads=1
 
 Please download silero_vad.onnx from
 https://github.com/snakers4/silero-vad/blob/master/files/silero_vad.onnx