diff --git a/paddle/operators/ctc_align_op.cc b/paddle/operators/ctc_align_op.cc
index 3fa8d2af7424ffc036d53b11c579b155ccabbe3a..eeecbd32127d2cf9756432817fc5d36673685aa7 100644
--- a/paddle/operators/ctc_align_op.cc
+++ b/paddle/operators/ctc_align_op.cc
@@ -89,4 +89,5 @@ namespace ops = paddle::operators;
 REGISTER_OPERATOR(ctc_align, ops::CTCAlignOp, ops::CTCAlignOpMaker,
                   paddle::framework::EmptyGradOpMaker);
 REGISTER_OP_CPU_KERNEL(
-    ctc_align, ops::CTCAlignKernel<paddle::platform::CPUDeviceContext, int>);
+    ctc_align, ops::CTCAlignKernel<paddle::platform::CPUDeviceContext, int>,
+    ops::CTCAlignKernel<paddle::platform::CPUDeviceContext, int64_t>);
diff --git a/paddle/operators/ctc_align_op.cu b/paddle/operators/ctc_align_op.cu
index 99e716e989f064193df46c486dcbf9348d6df8cb..45635f16745346b08f7e31db2f25905bdbc3aeeb 100644
--- a/paddle/operators/ctc_align_op.cu
+++ b/paddle/operators/ctc_align_op.cu
@@ -87,5 +87,5 @@ class CTCAlignOpCUDAKernel : public framework::OpKernel<T> {
 }  // namespace operators
 }  // namespace paddle
 
-REGISTER_OP_CUDA_KERNEL(ctc_align,
-                        paddle::operators::CTCAlignOpCUDAKernel<int>);
+REGISTER_OP_CUDA_KERNEL(ctc_align, paddle::operators::CTCAlignOpCUDAKernel<int>,
+                        paddle::operators::CTCAlignOpCUDAKernel<int64_t>);