diff options
| -rw-r--r-- | Makefile | 5 | ||||
| -rw-r--r-- | examples/tnet_preprocessing_example2.lua | 68 | ||||
| -rw-r--r-- | init.lua | 43 | ||||
| -rw-r--r-- | tnet_io/cwrapper.h | 1 | ||||
| -rw-r--r-- | tnet_io/test.c | 4 |
5 files changed, 118 insertions, 3 deletions
@@ -1,6 +1,9 @@ .PHONY: tnet INCLUDE := -I $(BUILD_DIR)/luajit-2.0/include/luajit-2.0 -all: tnet $(OBJ_DIR)/tnet_io/test $(LIB_DIR)/libspeech.so +all: tnet $(OBJ_DIR)/tnet_io/test $(LIB_DIR)/libspeech.so $(LUA_DIR)/speech/init.lua +$(LUA_DIR)/speech/init.lua: init.lua + mkdir -p $(LUA_DIR)/speech + cp $< $@ $(LIB_DIR)/libspeech.so: $(OBJ_DIR)/tnet_io/cwrapper.o $(OBJ_DIR)/init.o $(OBJ_DIR)/tnet_io/init.o $(OBJ_DIR)/tnet_io/libKaldiLib.a gcc -shared -o $@ $(OBJ_DIR)/tnet_io/cwrapper.o $(OBJ_DIR)/init.o $(OBJ_DIR)/tnet_io/libKaldiLib.a $(OBJ_DIR)/tnet_io/init.o -lstdc++ -Wl,-rpath=$(LIB_DIR) -L$(LIB_DIR) -lnerv $(OBJ_DIR)/tnet_io/test: $(OBJ_DIR)/tnet_io/cwrapper.o $(OBJ_DIR)/tnet_io/test.o $(OBJ_DIR)/tnet_io/libKaldiLib.a diff --git a/examples/tnet_preprocessing_example2.lua b/examples/tnet_preprocessing_example2.lua new file mode 100644 index 0000000..ae8d86b --- /dev/null +++ b/examples/tnet_preprocessing_example2.lua @@ -0,0 +1,68 @@ +require 'speech.init' +gconf = {mat_type = nerv.CuMatrixFloat, + batch_size = 158} +param_repo = nerv.ParamRepo({"global_transf.nerv"}) + +sublayer_repo = nerv.LayerRepo( + { + ["nerv.BiasLayer"] = + { + blayer1 = {{bias = "bias1"}, {dim_in = {429}, dim_out = {429}}}, + blayer2 = {{bias = "bias2"}, {dim_in = {429}, dim_out = {429}}} + }, + ["nerv.WindowLayer"] = + { + wlayer1 = {{window = "window1"}, {dim_in = {429}, dim_out = {429}}}, + wlayer2 = {{window = "window2"}, {dim_in = {429}, dim_out = {429}}} + } + }, param_repo, gconf) + +layer_repo = nerv.LayerRepo( + { + ["nerv.DAGLayer"] = + { + main = {{}, { + dim_in = {429}, dim_out = {429}, + sub_layers = sublayer_repo, + connections = { + ["<input>[1]"] = "blayer1[1]", + ["blayer1[1]"] = "wlayer1[1]", + ["wlayer1[1]"] = "blayer2[1]", + ["blayer2[1]"] = "wlayer2[1]", + ["wlayer2[1]"] = "<output>[1]" + } + }} + } + }, param_repo, gconf) + +reader = nerv.TNetReader({}, + { + id = "main_scp", + scp_file = "/slfs1/users/mfy43/swb_ivec/train_bp.scp", + conf_file = "/slfs1/users/mfy43/swb_ivec/plp_0_d_a.conf", + frm_ext = 5, + mlfs = { + ref = { + file = "/slfs1/users/mfy43/swb_ivec/ref.mlf", + format = "map", + format_arg = "/slfs1/users/mfy43/swb_ivec/dict", + dir = "*/", + ext = "lab" + } + }, + global_transf = layer_repo:get_layer("main") + }) + +utter = reader:get_data() +-- print(utter.main_scp) +print(utter.ref) +-- cf2 = nerv.ChunkFile("feat_256", "r") +-- input = cf2:read_chunk("input", gconf) + +-- for i = 0, 157 - 10 do +-- row_diff = input.trans[i] - utter.main_scp[i] +-- for j = 0, row_diff:ncol() - 1 do +-- nerv.utils.printf("%.8f ", row_diff[j]) +-- end +-- nerv.utils.printf("\n") +-- end diff --git a/init.lua b/init.lua new file mode 100644 index 0000000..1f20096 --- /dev/null +++ b/init.lua @@ -0,0 +1,43 @@ +require 'libspeech' +local TNetReader = nerv.class("nerv.TNetReader", "nerv.DataReader") + +function TNetReader:__init(global_conf, reader_conf) + self.feat_id = reader_conf.id + self.frm_ext = reader_conf.frm_ext + self.feat_repo = nerv.TNetFeatureRepo(reader_conf.scp_file, + reader_conf.conf_file, + reader_conf.frm_ext) + self.lab_repo = {} + for id, mlf_spec in pairs(reader_conf.mlfs) do + self.lab_repo[id] = nerv.TNetLabelRepo(mlf_spec.file, + mlf_spec.format, + mlf_spec.format_arg, + mlf_spec.dir, + mlf_spec.ext) + end + self.global_transf = reader_conf.global_transf +end + +function TNetReader:get_data() + local res = {} + local frm_ext = self.frm_ext + local step = frm_ext * 2 + 1 + local feat_utter = self.feat_repo:cur_utter() + local expanded = nerv.CuMatrixFloat(feat_utter:nrow(), feat_utter:ncol() * step) + expanded:expand_frm(nerv.CuMatrixFloat.new_from_host(feat_utter), frm_ext) + local rearranged = expanded:create() + rearranged:rearrange_frm(expanded, step) + local input = {rearranged} + local output = {rearranged:create()} + self.global_transf:init() + self.global_transf:propagate(input, output) + expanded = nerv.CuMatrixFloat(output[1]:nrow() - frm_ext * 2, output[1]:ncol()) + expanded:copy_fromd(output[1], frm_ext, feat_utter:nrow() - frm_ext) + res[self.feat_id] = expanded + for id, repo in pairs(self.lab_repo) do + local lab_utter = repo:get_utter(self.feat_repo, expanded:nrow()) + res[id] = lab_utter + end + self.feat_repo:next() + return res +end diff --git a/tnet_io/cwrapper.h b/tnet_io/cwrapper.h index 810901b..a34f090 100644 --- a/tnet_io/cwrapper.h +++ b/tnet_io/cwrapper.h @@ -1,6 +1,7 @@ #ifndef NERV_TNET_IO_CWRAPPER #define NERV_TNET_IO_CWRAPPER #include "../../matrix/generic/matrix.h" +#include "../../common.h" #ifdef __cplusplus extern "C" { #endif diff --git a/tnet_io/test.c b/tnet_io/test.c index a5b8f55..8c06805 100644 --- a/tnet_io/test.c +++ b/tnet_io/test.c @@ -22,7 +22,7 @@ int main() { "/slfs1/users/mfy43/swb_ivec/train_bp.scp", "/slfs1/users/mfy43/swb_ivec/plp_0_d_a.conf", 5); Matrix *feat_utter; - feat_utter = tnet_feature_repo_read_utterance(feat_repo); + feat_utter = tnet_feature_repo_read_utterance(feat_repo, NULL); TNetLabelRepo *lab_repo = tnet_label_repo_new( "/slfs1/users/mfy43/swb_ivec/ref.mlf", @@ -33,7 +33,7 @@ int main() { Matrix *lab_utter = tnet_label_repo_read_utterance(lab_repo, feat_utter->nrow - 5 * 2, tnet_feature_repo_current_samplerate(feat_repo), - tnet_feature_repo_current_tag(feat_repo)); + tnet_feature_repo_current_tag(feat_repo), NULL); print_nerv_matrix(lab_utter); return 0; } |