diff options
-rw-r--r-- | examples/tnet_preprocessing_example.lua | 2 | ||||
-rw-r--r-- | examples/tnet_preprocessing_example2.lua | 2 | ||||
-rw-r--r-- | examples/tnet_sgd_buffer.lua | 70 | ||||
-rw-r--r-- | init.lua | 16 | ||||
-rw-r--r-- | tnet_io/cwrapper.cpp | 6 | ||||
-rw-r--r-- | tnet_io/cwrapper.h | 1 | ||||
-rw-r--r-- | tnet_io/init.c | 7 | ||||
-rw-r--r-- | tools/tnet_to_nerv.c | 57 | ||||
-rw-r--r-- | tools/tnet_to_nerv.cpp | 68 |
9 files changed, 220 insertions, 9 deletions
diff --git a/examples/tnet_preprocessing_example.lua b/examples/tnet_preprocessing_example.lua index 8a65b44..5f20910 100644 --- a/examples/tnet_preprocessing_example.lua +++ b/examples/tnet_preprocessing_example.lua @@ -1,6 +1,6 @@ require 'libspeech' frm_ext = 5 -gconf = {mat_type = nerv.CuMatrixFloat, +gconf = {cumat_type = nerv.CuMatrixFloat, batch_size = 158} param_repo = nerv.ParamRepo({"global_transf.nerv"}) sublayer_repo = nerv.LayerRepo( diff --git a/examples/tnet_preprocessing_example2.lua b/examples/tnet_preprocessing_example2.lua index ae8d86b..8e1bc85 100644 --- a/examples/tnet_preprocessing_example2.lua +++ b/examples/tnet_preprocessing_example2.lua @@ -1,5 +1,5 @@ require 'speech.init' -gconf = {mat_type = nerv.CuMatrixFloat, +gconf = {cumat_type = nerv.CuMatrixFloat, batch_size = 158} param_repo = nerv.ParamRepo({"global_transf.nerv"}) diff --git a/examples/tnet_sgd_buffer.lua b/examples/tnet_sgd_buffer.lua new file mode 100644 index 0000000..152d2f5 --- /dev/null +++ b/examples/tnet_sgd_buffer.lua @@ -0,0 +1,70 @@ +require 'speech.init' +gconf = {cumat_type = nerv.CuMatrixFloat, + mmat_type = nerv.MMatrixFloat, + batch_size = 256} +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) + +tnet_reader = nerv.TNetReader({}, + { + id = "main_scp", +-- scp_file = "/slfs1/users/mfy43/swb_ivec/train_bp.scp", + scp_file = "t.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") + }) + +buffer = nerv.SGDBuffer(gconf, + { + buffer_size = 1024, + readers = { + { reader = tnet_reader, + data = {main_scp = 429, ref = 1}} + } + }) + +for data in buffer.get_data, buffer do + print(data.main_scp) +-- print(data.ref) +end @@ -4,6 +4,8 @@ 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.gconf = global_conf + self.global_transf = reader_conf.global_transf self.feat_repo = nerv.TNetFeatureRepo(reader_conf.scp_file, reader_conf.conf_file, reader_conf.frm_ext) @@ -15,24 +17,26 @@ function TNetReader:__init(global_conf, reader_conf) mlf_spec.dir, mlf_spec.ext) end - self.global_transf = reader_conf.global_transf end function TNetReader:get_data() + if self.feat_repo:is_end() then + return nil + end 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 expanded = self.gconf.cumat_type(feat_utter:nrow(), feat_utter:ncol() * step) + expanded:expand_frm(self.gconf.cumat_type.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:init(input[1]:nrow()) 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) + expanded = self.gconf.mmat_type(output[1]:nrow() - frm_ext * 2, output[1]:ncol()) + output[1]:copy_toh(expanded, 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()) diff --git a/tnet_io/cwrapper.cpp b/tnet_io/cwrapper.cpp index e82f3f8..4149557 100644 --- a/tnet_io/cwrapper.cpp +++ b/tnet_io/cwrapper.cpp @@ -2,13 +2,13 @@ #include "KaldiLib/Labels.h" #include "KaldiLib/Common.h" #include "KaldiLib/UserInterface.h" -#include "../../common.h" #include <string> #define SNAME "TNET" extern "C" { #include "cwrapper.h" #include "string.h" +#include "../../common.h" extern Matrix *nerv_matrix_host_float_new_(lua_State *L, long nrow, long ncol); @@ -77,6 +77,10 @@ extern "C" { repo->feature_repo.MoveNext(); } + int tnet_feature_repo_is_end(TNetFeatureRepo *repo) { + return repo->feature_repo.EndOfList(); + } + size_t tnet_feature_repo_current_samplerate(TNetFeatureRepo *repo) { return repo->feature_repo.CurrentHeader().mSamplePeriod; } diff --git a/tnet_io/cwrapper.h b/tnet_io/cwrapper.h index a34f090..54fb69b 100644 --- a/tnet_io/cwrapper.h +++ b/tnet_io/cwrapper.h @@ -14,6 +14,7 @@ extern "C" { size_t tnet_feature_repo_current_samplerate(TNetFeatureRepo *repo); const char *tnet_feature_repo_current_tag(TNetFeatureRepo *repo); void tnet_feature_repo_next(TNetFeatureRepo *repo); + int tnet_feature_repo_is_end(TNetFeatureRepo *repo); void tnet_feature_repo_destroy(TNetFeatureRepo *repo); typedef struct TNetLabelRepo TNetLabelRepo; diff --git a/tnet_io/init.c b/tnet_io/init.c index 3fa7cb8..16f6f37 100644 --- a/tnet_io/init.c +++ b/tnet_io/init.c @@ -40,10 +40,17 @@ static int feat_repo_next(lua_State *L) { return 0; } +static int feat_repo_is_end(lua_State *L) { + TNetFeatureRepo *repo = luaT_checkudata(L, 1, nerv_tnet_feat_repo_tname); + lua_pushboolean(L, tnet_feature_repo_is_end(repo)); + return 1; +} + static const luaL_Reg feat_repo_methods[] = { {"cur_utter", feat_repo_current_utterance}, {"cur_tag", feat_repo_current_tag}, {"next", feat_repo_next}, + {"is_end", feat_repo_is_end}, {NULL, NULL} }; diff --git a/tools/tnet_to_nerv.c b/tools/tnet_to_nerv.c new file mode 100644 index 0000000..f781236 --- /dev/null +++ b/tools/tnet_to_nerv.c @@ -0,0 +1,57 @@ +#include <stdio.h> +#include <string.h> +#include <stdlib.h> +char token[1024]; +double mat[4096][4096]; +int main() { + FILE *fout = fopen("converted.nerv", "w"); + int cnt = 0; + while (scanf("%s", token) != EOF) + { + int nrow, ncol; + int i, j; + if (strcmp(token, "<biasedlinearity>") == 0) + { + scanf("%d %d", &ncol, &nrow); + scanf("%s %d %d", token, &ncol, &nrow); + printf("%d %d\n", nrow, ncol); + for (j = 0; j < ncol; j++) + for (i = 0; i < nrow; i++) + scanf("%lf", mat[i] + j); + off_t base = ftello(fout); + fprintf(fout, "%16d", 0); + fprintf(fout, "{type=\"nerv.LinearTransParam\",id=\"affine%d_ltp\"}\n", + cnt); + fprintf(fout, "%d %d\n", nrow, ncol); + for (i = 0; i < nrow; i++) + { + for (j = 0; j < ncol; j++) + fprintf(fout, "%.8f ", mat[i][j]); + fprintf(fout, "\n"); + } + size_t length = ftello(fout) - base; + fseeko(fout, base, SEEK_SET); + fprintf(fout, "[%13lu]\n", length); + fseeko(fout, 0, SEEK_END); + if (scanf("%s %d", token, &ncol) == 2 && *token == 'v') + { + base = ftello(fout); + for (j = 0; j < ncol; j++) + scanf("%lf", mat[0] + j); + fprintf(fout, "%16d", 0); + fprintf(fout, "{type=\"nerv.BiasParam\",id=\"affine%d_bp\"}\n", + cnt); + fprintf(fout, "1 %d\n", nrow, ncol); + for (j = 0; j < ncol; j++) + fprintf(fout, "%.8f ", mat[0][j]); + fprintf(fout, "\n"); + length = ftello(fout) - base; + fseeko(fout, base, SEEK_SET); + fprintf(fout, "[%13lu]\n", length); + cnt++; + fseeko(fout, 0, SEEK_END); + } + } + } + return 0; +} diff --git a/tools/tnet_to_nerv.cpp b/tools/tnet_to_nerv.cpp new file mode 100644 index 0000000..cedf27a --- /dev/null +++ b/tools/tnet_to_nerv.cpp @@ -0,0 +1,68 @@ +#include <cstdio> +#include <fstream> +#include <string> +#include <cstring> +char token[1024]; +char output[1024]; +double mat[4096][4096]; +int main() { + std::ofstream fout; + fout.open("converted.nerv"); + int cnt = 0; + while (scanf("%s", token) != EOF) + { + int nrow, ncol; + int i, j; + if (strcmp(token, "<biasedlinearity>") == 0) + { + scanf("%d %d", &ncol, &nrow); + scanf("%s %d %d", token, &ncol, &nrow); + printf("%d %d\n", nrow, ncol); + for (j = 0; j < ncol; j++) + for (i = 0; i < nrow; i++) + scanf("%lf", mat[i] + j); + long base = fout.tellp(); + sprintf(output, "%16d", 0); + fout << output; + sprintf(output, "{type=\"nerv.LinearTransParam\",id=\"affine%d_ltp\"}\n", + cnt); + fout << output; + sprintf(output, "%d %d\n", nrow, ncol); + fout << output; + for (i = 0; i < nrow; i++) + { + for (j = 0; j < ncol; j++) + fout << mat[i][j] << " "; + fout << std::endl; + } + long length = fout.tellp() - base; + fout.seekp(base); + sprintf(output, "[%13lu]\n", length); + fout << output; + fout.seekp(0, std::ios_base::end); + if (scanf("%s %d", token, &ncol) == 2 && *token == 'v') + { + base = fout.tellp(); + for (j = 0; j < ncol; j++) + scanf("%lf", mat[0] + j); + sprintf(output, "%16d", 0); + fout << output; + sprintf(output, "{type=\"nerv.BiasParam\",id=\"affine%d_bp\"}\n", + cnt); + fout << output; + sprintf(output, "1 %d\n", ncol); + fout << output; + for (j = 0; j < ncol; j++) + fout << mat[0][j] << " "; + fout << std::endl; + length = fout.tellp() - base; + fout.seekp(base); + sprintf(output, "[%13lu]\n", length); + fout << output; + fout.seekp(0, std::ios_base::end); + cnt++; + } + } + } + return 0; +} |