diff --git a/.github/workflows/ubuntu-cuda-wheels.yml b/.github/workflows/ubuntu-cuda-wheels.yml
index bd4c006..4e48695 100644
--- a/.github/workflows/ubuntu-cuda-wheels.yml
+++ b/.github/workflows/ubuntu-cuda-wheels.yml
@@ -65,7 +65,7 @@ jobs:
         uses: addnab/docker-run-action@v3
         with:
             image: ${{ matrix.image }}
-            options: -v ${{ github.workspace }}:/var/www -e PYTHON_VERSION=${{ matrix.python-version }} -e TORCH_VERSION=${{ matrix.torch }} -e CUDA_VERSION=${{ matrix.cuda }}
+            options: -v ${{ github.workspace }}:/var/www -e IS_2_28=${{ matrix.is_2_28 }} -e PYTHON_VERSION=${{ matrix.python-version }} -e TORCH_VERSION=${{ matrix.torch }} -e CUDA_VERSION=${{ matrix.cuda }}
             run: |
               echo "pwd: $PWD"
               uname -a
@@ -81,6 +81,42 @@ jobs:
 
               ls -lh /opt/python/*/bin
 
+              # cp36-cp36m
+              # cp37-cp37m
+              # cp38-cp38
+              # cp39-cp39
+              # cp310-cp310
+              # cp311-cp311
+              # cp312-cp312
+              # cp313-cp313
+              # cp313-cp313t  (no gil)
+              if [[ $PYTHON_VERSION == "3.6" ]]; then
+                python_dir=/opt/python/cp36-cp36m
+              elif [[ $PYTHON_VERSION == "3.7" ]]; then
+                python_dir=/opt/python/cp37-cp37m
+              elif [[ $PYTHON_VERSION == "3.8" ]]; then
+                python_dir=/opt/python/cp38-cp38
+              elif [[ $PYTHON_VERSION == "3.9" ]]; then
+                python_dir=/opt/python/cp39-cp39
+              elif [[ $PYTHON_VERSION == "3.10" ]]; then
+                python_dir=/opt/python/cp310-cp310
+              elif [[ $PYTHON_VERSION == "3.11" ]]; then
+                python_dir=/opt/python/cp311-cp311
+              elif [[ $PYTHON_VERSION == "3.12" ]]; then
+                python_dir=/opt/python/cp312-cp312
+              elif [[ $PYTHON_VERSION == "3.13" ]]; then
+                python_dir=/opt/python/cp313-cp313
+              else
+                echo "Unsupported Python version $PYTHON_VERSION"
+                exit 1
+              fi
+
+              export PYTHON_INSTALL_DIR=$python_dir
+              export PATH=$PYTHON_INSTALL_DIR/bin:$PATH
+
+              python3 --version
+              which python3
+
               pushd /usr/local
               rm cuda
               ln -s cuda-$CUDA_VERSION cuda
diff --git a/scripts/github_actions/build-ubuntu-cuda.sh b/scripts/github_actions/build-ubuntu-cuda.sh
index 4f2451f..018c8cd 100755
--- a/scripts/github_actions/build-ubuntu-cuda.sh
+++ b/scripts/github_actions/build-ubuntu-cuda.sh
@@ -101,9 +101,17 @@ export CMAKE_CUDA_COMPILER_LAUNCHER=
 export KALDIFEAT_CMAKE_ARGS=" -DPYTHON_EXECUTABLE=$PYTHON_INSTALL_DIR/bin/python3 "
 export KALDIFEAT_MAKE_ARGS=" -j2 "
 
+echo "KALDIFEAT_CMAKE_ARGS: $KALDIFEAT_CMAKE_ARGS"
+
 
 python3 setup.py bdist_wheel
 
+if [[ x"$IS_2_28" == x"1" ]]; then
+  plat=manylinux_2_28_x86_64
+else
+  plat=manylinux_2_17_x86_64
+fi
+
 auditwheel --verbose repair \
   --exclude libc10.so \
   --exclude libc10_cuda.so \
@@ -143,7 +151,7 @@ auditwheel --verbose repair \
   --exclude libshm.so \
   --exclude libtorch_cuda_cpp.so \
   --exclude libtorch_cuda_cu.so \
-  --plat manylinux_2_17_x86_64 \
+  --plat $plat \
   -w /var/www/wheelhouse \
   dist/*.whl
 
diff --git a/scripts/github_actions/generate_build_matrix.py b/scripts/github_actions/generate_build_matrix.py
index 4994da0..f9cd94a 100755
--- a/scripts/github_actions/generate_build_matrix.py
+++ b/scripts/github_actions/generate_build_matrix.py
@@ -339,6 +339,7 @@ def generate_build_matrix(enable_cuda, for_windows, for_macos, test_only_latest_
                                 "python-version": p,
                                 "cuda": c,
                                 "image": f"pytorch/manylinux2_28-builder:cuda{c}",
+                                "is_2_28": "1",
                             }
                         )
                         continue
@@ -349,6 +350,7 @@ def generate_build_matrix(enable_cuda, for_windows, for_macos, test_only_latest_
                             "python-version": p,
                             "cuda": c,
                             "image": f"pytorch/manylinux-builder:cuda{c}",
+                            "is_2_28": "0",
                         }
                     )
         else:
@@ -368,6 +370,7 @@ def generate_build_matrix(enable_cuda, for_windows, for_macos, test_only_latest_
                             "python-version": p,
                             #  "image": "pytorch/manylinux-builder:cpu-2.4",
                             "image": "pytorch/manylinux-builder:cpu-27677ead7c8293c299a885ae2c474bf445e653a5",
+                            "is_2_28": "0",
                         }
                     )
                 elif version_ge(torch, "2.2.0"):
@@ -376,6 +379,7 @@ def generate_build_matrix(enable_cuda, for_windows, for_macos, test_only_latest_
                             "torch": torch,
                             "python-version": p,
                             "image": "pytorch/manylinux-builder:cpu-2.2",
+                            "is_2_28": "0",
                         }
                     )
                 else:
@@ -384,6 +388,7 @@ def generate_build_matrix(enable_cuda, for_windows, for_macos, test_only_latest_
                             "torch": torch,
                             "python-version": p,
                             "image": f"pytorch/manylinux-builder:cuda10.2",
+                            "is_2_28": "0",
                         }
                     )