fix as comment

speechx/CMakeLists.txt

@@ -102,7 +102,16 @@ message(STATUS "Pybind11_INCLUDES = ${pybind11_INCLUDE_DIRS}, pybind11_LIBRARIES
 # paddle include and link option
 # -L/workspace/DeepSpeech-2.x/speechx/venv/lib/python3.7/site-packages/paddle/libs -L/workspace/DeepSpeech-2.x/speechx/venv/lib/python3.7/site-packages/paddle/fluid -l:libpaddle.so -l:libdnnl.so.2 -l:libiomp5.so
 execute_process(
-    COMMAND python -c "import os; import paddle; include_dir=paddle.sysconfig.get_include(); paddle_dir=os.path.split(include_dir)[0]; libs_dir=os.path.join(paddle_dir, 'libs'); fluid_dir=os.path.join(paddle_dir, 'fluid'); out=' '.join([\"-L\" + libs_dir, \"-L\" + fluid_dir]); out += \" -l:libpaddle.so -l:libdnnl.so.2 -l:libiomp5.so\"; print(out);"
+    COMMAND python -c "\
+import os;\
+import paddle;\
+include_dir=paddle.sysconfig.get_include();\
+paddle_dir=os.path.split(include_dir)[0];\
+libs_dir=os.path.join(paddle_dir, 'libs');\
+fluid_dir=os.path.join(paddle_dir, 'fluid');\
+out=' '.join([\"-L\" + libs_dir, \"-L\" + fluid_dir]);\
+out += \" -l:libpaddle.so -l:libdnnl.so.2 -l:libiomp5.so\"; print(out);\
+    "
     OUTPUT_VARIABLE PADDLE_LINK_FLAGS
     RESULT_VARIABLE SUCESS)
 
@@ -112,7 +121,11 @@ string(STRIP ${PADDLE_LINK_FLAGS} PADDLE_LINK_FLAGS)
 # paddle compile option
 # -I/workspace/DeepSpeech-2.x/speechx/venv/lib/python3.7/site-packages/paddle/include
 execute_process(
-    COMMAND python -c "import paddle; include_dir = paddle.sysconfig.get_include(); print(f\"-I{include_dir}\");"
+    COMMAND python -c "\
+import paddle; \
+include_dir = paddle.sysconfig.get_include(); \
+print(f\"-I{include_dir}\"); \
+    "
     OUTPUT_VARIABLE PADDLE_COMPILE_FLAGS)
 message(STATUS PADDLE_COMPILE_FLAGS= ${PADDLE_COMPILE_FLAGS})
 string(STRIP ${PADDLE_COMPILE_FLAGS} PADDLE_COMPILE_FLAGS)
@@ -121,7 +134,15 @@ string(STRIP ${PADDLE_COMPILE_FLAGS} PADDLE_COMPILE_FLAGS)
 # for LD_LIBRARY_PATH
 # set(PADDLE_LIB_DIRS /workspace/DeepSpeech-2.x/tools/venv/lib/python3.7/site-packages/paddle/fluid:/workspace/DeepSpeech-2.x/tools/venv/lib/python3.7/site-packages/paddle/libs/)
 execute_process(
-    COMMAND python -c "import os; import paddle; include_dir=paddle.sysconfig.get_include(); paddle_dir=os.path.split(include_dir)[0]; libs_dir=os.path.join(paddle_dir, 'libs'); fluid_dir=os.path.join(paddle_dir, 'fluid'); out=':'.join([libs_dir, fluid_dir]); print(out);"
+    COMMAND python -c " \
+import os; \
+import paddle; \
+include_dir=paddle.sysconfig.get_include(); \
+paddle_dir=os.path.split(include_dir)[0]; \
+libs_dir=os.path.join(paddle_dir, 'libs'); \
+fluid_dir=os.path.join(paddle_dir, 'fluid'); \
+out=':'.join([libs_dir, fluid_dir]); print(out); \
+    "
     OUTPUT_VARIABLE PADDLE_LIB_DIRS)
 message(STATUS PADDLE_LIB_DIRS= ${PADDLE_LIB_DIRS})
 

speechx/README.md

@@ -9,7 +9,7 @@ We develop under:
 * gcc/g++/gfortran - 8.2.0
 * cmake - 3.16.0
 
-> Please using `tools/env.sh` to create python `venv`, then `source venv/bin/activate` to build speechx.
+> Please use `tools/env.sh` to create python `venv`, then `source venv/bin/activate` to build speechx.
 
 > We make sure all things work fun under docker, and recommend using it to develop and deploy.
 
@@ -35,7 +35,7 @@ bash tools/venv.sh
 
 2. Build `speechx` and `examples`.
 
-For now we using feature under `develop` branch of paddle, so we need install `paddlepaddle` nightly build version.
+For now we are using feature under `develop` branch of paddle, so we need to install `paddlepaddle` nightly build version.
 For example: 
 ```
 source venv/bin/activate

speechx/cmake/paddleinference.cmake

@@ -1,14 +1,5 @@
 set(paddle_SOURCE_DIR ${fc_patch}/paddle-lib)
 set(paddle_PREFIX_DIR ${fc_patch}/paddle-lib-prefix)
-# ExternalProject_Add(paddle
-#   URL      https://paddle-inference-lib.bj.bcebos.com/2.2.2/cxx_c/Linux/CPU/gcc8.2_avx_mkl/paddle_inference.tgz
-#   URL_HASH SHA256=7c6399e778c6554a929b5a39ba2175e702e115145e8fa690d2af974101d98873
-#   PREFIX            ${paddle_PREFIX_DIR} 
-#   SOURCE_DIR        ${paddle_SOURCE_DIR}
-#   CONFIGURE_COMMAND ""
-#   BUILD_COMMAND     ""
-#   INSTALL_COMMAND   ""
-# )
 
 include(FetchContent)
 FetchContent_Declare(

speechx/examples/codelab/feat/run.sh

@@ -42,8 +42,8 @@ mkdir -p $exp_dir
 export GLOG_logtostderr=1
 
 cmvn_json2kaldi_main \
-    --json_file  $model_dir/data/mean_std.json \
-    --cmvn_write_path $exp_dir/cmvn.ark \
+    --json_file=$model_dir/data/mean_std.json \
+    --cmvn_write_path=$exp_dir/cmvn.ark \
     --binary=false
 echo "convert json cmvn to kaldi ark."
 
@@ -55,7 +55,7 @@ compute_linear_spectrogram_main \
 echo "compute linear spectrogram feature."
 
 compute_fbank_main \
-    --num_bins 161 \
+    --num_bins=161 \
     --wav_rspecifier=scp:$data_dir/wav.scp \
     --feature_wspecifier=ark,t:$exp_dir/fbank.ark \
     --cmvn_file=$exp_dir/cmvn.ark

speechx/examples/codelab/u2/local/decode.sh

@@ -7,7 +7,7 @@ set -e
 data=data
 exp=exp
 mkdir -p $exp
-ckpt_dir=./data/model
+ckpt_dir=$data/model
 model_dir=$ckpt_dir/asr1_chunk_conformer_u2pp_wenetspeech_static_1.1.0.model/
 
 ctc_prefix_beam_search_decoder_main \

speechx/speechx/decoder/CMakeLists.txt

@@ -4,11 +4,11 @@ set(srcs)
 
 if (USING_DS2)
 list(APPEND srcs
-ctc_decoders/decoder_utils.cpp
-ctc_decoders/path_trie.cpp
-ctc_decoders/scorer.cpp
-ctc_beam_search_decoder.cc
-ctc_tlg_decoder.cc
+  ctc_decoders/decoder_utils.cpp
+  ctc_decoders/path_trie.cpp
+  ctc_decoders/scorer.cpp
+  ctc_beam_search_decoder.cc
+  ctc_tlg_decoder.cc
 )
 endif()
 

speechx/speechx/decoder/ctc_beam_search_opt.h

@@ -11,12 +11,11 @@
 // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 // See the License for the specific language governing permissions and
 // limitations under the License.
+#pragma once
 
 #include "base/common.h"
 #include "util/parse-options.h"
 
-#pragma once
-
 namespace ppspeech {
 
 

speechx/speechx/decoder/ctc_prefix_beam_search_decoder.cc

@@ -107,12 +107,12 @@ void CTCPrefixBeamSearch::AdvanceDecoding(
         std::min(static_cast<int>(logp[0].size()), opts_.first_beam_size);
 
     for (int t = 0; t < logp.size(); ++t, ++num_frame_decoded_) {
-        const std::vector<float>& logp_t = logp[t];
+        const std::vector<kaldi::BaseFloat>& logp_t = logp[t];
         std::unordered_map<std::vector<int>, PrefixScore, PrefixScoreHash>
             next_hyps;
 
         // 1. first beam prune, only select topk candidates
-        std::vector<float> topk_score;
+        std::vector<kaldi::BaseFloat> topk_score;
         std::vector<int32_t> topk_index;
         TopK(logp_t, first_beam_size, &topk_score, &topk_index);
         VLOG(2) << "topk: " << num_frame_decoded_ << " "

speechx/speechx/decoder/ctc_prefix_beam_search_decoder.h

@@ -27,7 +27,7 @@ namespace ppspeech {
 class ContextGraph;
 class CTCPrefixBeamSearch : public DecoderBase {
   public:
-    explicit CTCPrefixBeamSearch(const std::string& vocab_path,
+    CTCPrefixBeamSearch(const std::string& vocab_path,
                                  const CTCBeamSearchOptions& opts);
     ~CTCPrefixBeamSearch() {}