Yang Zhou
2 years ago
parent
6f0b3a15d7
commit
143ab13679
@ -1,5 +1,14 @@
|
||||
cmake_minimum_required(VERSION 3.14 FATAL_ERROR)
|
||||
|
||||
add_executable(offline_decoder_sliding_chunk_main ${CMAKE_CURRENT_SOURCE_DIR}/offline_decoder_sliding_chunk_main.cc)
|
||||
target_include_directories(offline_decoder_sliding_chunk_main PRIVATE ${SPEECHX_ROOT} ${SPEECHX_ROOT}/kaldi)
|
||||
target_link_libraries(offline_decoder_sliding_chunk_main PUBLIC nnet decoder fst utils gflags glog kaldi-base kaldi-matrix kaldi-util ${DEPS})
|
||||
|
||||
add_executable(offline_decoder_main ${CMAKE_CURRENT_SOURCE_DIR}/offline_decoder_main.cc)
|
||||
target_include_directories(offline_decoder_main PRIVATE ${SPEECHX_ROOT} ${SPEECHX_ROOT}/kaldi)
|
||||
target_link_libraries(offline_decoder_main PUBLIC nnet decoder fst utils gflags glog kaldi-base kaldi-matrix kaldi-util ${DEPS})
|
||||
|
||||
add_executable(decoder_test_main ${CMAKE_CURRENT_SOURCE_DIR}/decoder_test_main.cc)
|
||||
target_include_directories(decoder_test_main PRIVATE ${SPEECHX_ROOT} ${SPEECHX_ROOT}/kaldi)
|
||||
target_link_libraries(decoder_test_main PUBLIC nnet decoder fst utils gflags glog kaldi-base kaldi-matrix kaldi-util ${DEPS})
|
||||
|
||||
|
||||
@ -0,0 +1,69 @@
|
||||
// Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
|
||||
//
|
||||
// Licensed under the Apache License, Version 2.0 (the "License");
|
||||
// you may not use this file except in compliance with the License.
|
||||
// You may obtain a copy of the License at
|
||||
//
|
||||
// http://www.apache.org/licenses/LICENSE-2.0
|
||||
//
|
||||
// Unless required by applicable law or agreed to in writing, software
|
||||
// distributed under the License is distributed on an "AS IS" BASIS,
|
||||
// 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.
|
||||
|
||||
// todo refactor, repalce with gtest
|
||||
|
||||
#include "base/flags.h"
|
||||
#include "base/log.h"
|
||||
#include "decoder/ctc_beam_search_decoder.h"
|
||||
#include "kaldi/util/table-types.h"
|
||||
#include "nnet/decodable.h"
|
||||
|
||||
DEFINE_string(nnet_prob_respecifier, "", "test nnet prob rspecifier");
|
||||
DEFINE_string(dict_file, "vocab.txt", "vocabulary of lm");
|
||||
DEFINE_string(lm_path, "lm.klm", "language model");
|
||||
|
||||
|
||||
using kaldi::BaseFloat;
|
||||
using kaldi::Matrix;
|
||||
using std::vector;
|
||||
|
||||
int main(int argc, char* argv[]) {
|
||||
gflags::ParseCommandLineFlags(&argc, &argv, false);
|
||||
google::InitGoogleLogging(argv[0]);
|
||||
|
||||
kaldi::SequentialBaseFloatMatrixReader likelihood_reader(
|
||||
FLAGS_nnet_prob_respecifier);
|
||||
std::string dict_file = FLAGS_dict_file;
|
||||
std::string lm_path = FLAGS_lm_path;
|
||||
|
||||
int32 num_done = 0, num_err = 0;
|
||||
|
||||
ppspeech::CTCBeamSearchOptions opts;
|
||||
opts.dict_file = dict_file;
|
||||
opts.lm_path = lm_path;
|
||||
ppspeech::CTCBeamSearch decoder(opts);
|
||||
|
||||
std::shared_ptr<ppspeech::Decodable> decodable(
|
||||
new ppspeech::Decodable(nullptr, nullptr));
|
||||
|
||||
decoder.InitDecoder();
|
||||
|
||||
for (; !likelihood_reader.Done(); likelihood_reader.Next()) {
|
||||
string utt = likelihood_reader.Key();
|
||||
const kaldi::Matrix<BaseFloat> likelihood = likelihood_reader.Value();
|
||||
decodable->Acceptlikelihood(likelihood);
|
||||
decoder.AdvanceDecode(decodable);
|
||||
std::string result;
|
||||
result = decoder.GetFinalBestPath();
|
||||
KALDI_LOG << " the result of " << utt << " is " << result;
|
||||
decodable->Reset();
|
||||
decoder.Reset();
|
||||
++num_done;
|
||||
}
|
||||
|
||||
KALDI_LOG << "Done " << num_done << " utterances, " << num_err
|
||||
<< " with errors.";
|
||||
return (num_done != 0 ? 0 : 1);
|
||||
}
|
||||
@ -0,0 +1,119 @@
|
||||
// Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
|
||||
//
|
||||
// Licensed under the Apache License, Version 2.0 (the "License");
|
||||
// you may not use this file except in compliance with the License.
|
||||
// You may obtain a copy of the License at
|
||||
//
|
||||
// http://www.apache.org/licenses/LICENSE-2.0
|
||||
//
|
||||
// Unless required by applicable law or agreed to in writing, software
|
||||
// distributed under the License is distributed on an "AS IS" BASIS,
|
||||
// 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.
|
||||
|
||||
// todo refactor, repalce with gtest
|
||||
|
||||
#include "base/flags.h"
|
||||
#include "base/log.h"
|
||||
#include "decoder/ctc_beam_search_decoder.h"
|
||||
#include "frontend/raw_audio.h"
|
||||
#include "kaldi/util/table-types.h"
|
||||
#include "nnet/decodable.h"
|
||||
#include "nnet/paddle_nnet.h"
|
||||
|
||||
DEFINE_string(feature_respecifier, "", "test feature rspecifier");
|
||||
DEFINE_string(model_path, "avg_1.jit.pdmodel", "paddle nnet model");
|
||||
DEFINE_string(param_path, "avg_1.jit.pdiparams", "paddle nnet model param");
|
||||
DEFINE_string(dict_file, "vocab.txt", "vocabulary of lm");
|
||||
DEFINE_string(lm_path, "lm.klm", "language model");
|
||||
|
||||
|
||||
using kaldi::BaseFloat;
|
||||
using kaldi::Matrix;
|
||||
using std::vector;
|
||||
|
||||
int main(int argc, char* argv[]) {
|
||||
gflags::ParseCommandLineFlags(&argc, &argv, false);
|
||||
google::InitGoogleLogging(argv[0]);
|
||||
|
||||
kaldi::SequentialBaseFloatMatrixReader feature_reader(
|
||||
FLAGS_feature_respecifier);
|
||||
std::string model_graph = FLAGS_model_path;
|
||||
std::string model_params = FLAGS_param_path;
|
||||
std::string dict_file = FLAGS_dict_file;
|
||||
std::string lm_path = FLAGS_lm_path;
|
||||
|
||||
int32 num_done = 0, num_err = 0;
|
||||
|
||||
ppspeech::CTCBeamSearchOptions opts;
|
||||
opts.dict_file = dict_file;
|
||||
opts.lm_path = lm_path;
|
||||
ppspeech::CTCBeamSearch decoder(opts);
|
||||
|
||||
ppspeech::ModelOptions model_opts;
|
||||
model_opts.model_path = model_graph;
|
||||
model_opts.params_path = model_params;
|
||||
model_opts.cache_shape = "5-1-1024,5-1-1024";
|
||||
std::shared_ptr<ppspeech::PaddleNnet> nnet(
|
||||
new ppspeech::PaddleNnet(model_opts));
|
||||
std::shared_ptr<ppspeech::RawDataCache> raw_data(
|
||||
new ppspeech::RawDataCache());
|
||||
std::shared_ptr<ppspeech::Decodable> decodable(
|
||||
new ppspeech::Decodable(nnet, raw_data));
|
||||
|
||||
int32 chunk_size = 7;
|
||||
int32 chunk_stride = 4;
|
||||
int32 receptive_field_length = 7;
|
||||
decoder.InitDecoder();
|
||||
|
||||
for (; !feature_reader.Done(); feature_reader.Next()) {
|
||||
string utt = feature_reader.Key();
|
||||
kaldi::Matrix<BaseFloat> feature = feature_reader.Value();
|
||||
raw_data->SetDim(feature.NumCols());
|
||||
int32 row_idx = 0;
|
||||
int32 padding_len = 0;
|
||||
int32 ori_feature_len = feature.NumRows();
|
||||
if ( (feature.NumRows() - chunk_size) % chunk_stride != 0) {
|
||||
padding_len = chunk_stride - (feature.NumRows() - chunk_size) % chunk_stride;
|
||||
feature.Resize(feature.NumRows() + padding_len, feature.NumCols(), kaldi::kCopyData);
|
||||
}
|
||||
int32 num_chunks = (feature.NumRows() - chunk_size) / chunk_stride + 1;
|
||||
for (int chunk_idx = 0; chunk_idx < num_chunks; ++chunk_idx) {
|
||||
kaldi::Vector<kaldi::BaseFloat> feature_chunk(chunk_size *
|
||||
feature.NumCols());
|
||||
int32 feature_chunk_size = 0;
|
||||
if ( ori_feature_len > chunk_idx * chunk_stride) {
|
||||
feature_chunk_size = std::min(ori_feature_len - chunk_idx * chunk_stride, chunk_size);
|
||||
}
|
||||
if (feature_chunk_size < receptive_field_length) break;
|
||||
|
||||
int32 start = chunk_idx * chunk_stride;
|
||||
int32 end = start + chunk_size;
|
||||
|
||||
for (int row_id = 0; row_id < chunk_size; ++row_id) {
|
||||
kaldi::SubVector<kaldi::BaseFloat> tmp(feature, start);
|
||||
kaldi::SubVector<kaldi::BaseFloat> f_chunk_tmp(
|
||||
feature_chunk.Data() + row_id * feature.NumCols(),
|
||||
feature.NumCols());
|
||||
f_chunk_tmp.CopyFromVec(tmp);
|
||||
++start;
|
||||
}
|
||||
raw_data->Accept(feature_chunk);
|
||||
if (chunk_idx == num_chunks - 1) {
|
||||
raw_data->SetFinished();
|
||||
}
|
||||
decoder.AdvanceDecode(decodable);
|
||||
}
|
||||
std::string result;
|
||||
result = decoder.GetFinalBestPath();
|
||||
KALDI_LOG << " the result of " << utt << " is " << result;
|
||||
decodable->Reset();
|
||||
decoder.Reset();
|
||||
++num_done;
|
||||
}
|
||||
|
||||
KALDI_LOG << "Done " << num_done << " utterances, " << num_err
|
||||
<< " with errors.";
|
||||
return (num_done != 0 ? 0 : 1);
|
||||
}
|
||||
Loading…
Reference in new issue