Add KalpyAligner wrapper class

.pre-commit-config.yaml

@@ -36,4 +36,5 @@ repos:
       - id: trailing-whitespace
       - id: check-added-large-files
         args: ['--maxkb=2000']
+        exclude: 'tests/data'
       - id: mixed-line-ending

docs/source/conf.py

@@ -130,7 +130,6 @@
         "Corpus Phonetics Tutorial",
         "https://eleanorchodroff.com/tutorial/intro.html",
     ),
-    "coqui": ("Coqui", "https://coqui.ai/"),
     "conda_installation": (
         "Conda installation",
         "https://docs.conda.io/projects/conda/en/latest/user-guide/install/index.html",

environment.yml

@@ -16,6 +16,7 @@ dependencies:
   - pytest
   - pip
   - dataclassy
+  - pyyaml
   - praatio
   - pytest-order
   - pip:

extensions/gmm/gmm.cpp

@@ -2600,4 +2600,92 @@ void init_gmm(py::module &_m) {
         py::arg("features"),
         py::return_value_policy::take_ownership
         );
+
+  m.def("gmm_interpolate_boundary_fast",
+        [](const AmDiagGmm &am_gmm,
+          const TransitionModel &trans_model,
+          const Matrix<BaseFloat> &features,
+          const int32 previous_transition_id,
+          const int32 following_transition_id){
+
+          py::gil_scoped_release release;
+          std::vector<BaseFloat> ratios;
+          BaseFloat previous_phone_log_likelihood, following_phone_log_likelihood;
+          int32 previous_pdf_id = trans_model.TransitionIdToPdfFast(previous_transition_id);
+          int32 following_pdf_id = trans_model.TransitionIdToPdfFast(following_transition_id);
+          BaseFloat min_ratio, max_ratio, ratio;
+          min_ratio = 100000.0;
+          max_ratio = -10000.0;
+          for (int32 i = 0; i < features.NumRows(); i++) {
+            SubVector<BaseFloat> feat_row(features, i);
+            previous_phone_log_likelihood = am_gmm.LogLikelihood(previous_pdf_id, feat_row);
+            following_phone_log_likelihood = am_gmm.LogLikelihood(following_pdf_id, feat_row);
+            ratio = previous_phone_log_likelihood - following_phone_log_likelihood;
+            ratios.push_back(ratio);
+            if (ratio > max_ratio)
+              max_ratio = ratio;
+            if (ratio < min_ratio)
+              min_ratio = ratio;
+
+          }
+          for (int32 j = 0; j < ratios.size(); j++) {
+            ratio = ratios[j];
+            if (-1 + (2*(ratio - min_ratio) / (max_ratio - min_ratio)) < 0.0)
+              return j;
+          }
+          return -1;
+                       },
+        py::arg("am_gmm"),
+        py::arg("trans_model"),
+        py::arg("features"),
+        py::arg("previous_transition_id"),
+        py::arg("following_transition_id"),
+        py::return_value_policy::take_ownership
+        );
+
+  m.def("gmm_interpolate_boundary",
+        [](const AmDiagGmm &am_gmm,
+          const TransitionModel &trans_model,
+          const Matrix<BaseFloat> &features,
+          const int32 previous_transition_id,
+          const int32 following_transition_id){
+
+          py::gil_scoped_release release;
+          std::vector<BaseFloat> ratios, normalized_ratios;
+          BaseFloat previous_phone_log_likelihood, following_phone_log_likelihood;
+          int32 previous_pdf_id = trans_model.TransitionIdToPdfFast(previous_transition_id);
+          int32 following_pdf_id = trans_model.TransitionIdToPdfFast(following_transition_id);
+          BaseFloat min_ratio, max_ratio, ratio, normalized_ratio;
+          min_ratio = 100000.0;
+          max_ratio = -10000.0;
+          for (int32 i = 0; i < features.NumRows(); i++) {
+            SubVector<BaseFloat> feat_row(features, i);
+            previous_phone_log_likelihood = am_gmm.LogLikelihood(previous_pdf_id, feat_row);
+            following_phone_log_likelihood = am_gmm.LogLikelihood(following_pdf_id, feat_row);
+            ratio = previous_phone_log_likelihood - following_phone_log_likelihood;
+            ratios.push_back(ratio);
+            if (ratio > max_ratio)
+              max_ratio = ratio;
+            if (ratio < min_ratio)
+              min_ratio = ratio;
+
+          }
+          int32 boundary_index = -1;
+          for (int32 j = 0; j < ratios.size(); j++) {
+            ratio = ratios[j];
+            normalized_ratio = -1 + (2*(ratio - min_ratio) / (max_ratio - min_ratio));
+            if (normalized_ratio < 0.0)
+              boundary_index = j;
+            normalized_ratios.push_back(normalized_ratio);
+          }
+          py::gil_scoped_acquire acquire;
+          return py::make_tuple(boundary_index, normalized_ratios);
+                       },
+        py::arg("am_gmm"),
+        py::arg("trans_model"),
+        py::arg("features"),
+        py::arg("previous_transition_id"),
+        py::arg("following_transition_id"),
+        py::return_value_policy::take_ownership
+        );
 }

extensions/matrix/matrix.cpp

@@ -806,6 +806,9 @@ void pybind_kaldi_vector(py::module& m) {
     ));
 
   py::class_<SubVector<float>, VectorBase<float>>(m, "FloatSubVector")
+      .def(py::init<const MatrixBase<float> &, MatrixIndexT>(),
+           py::arg("matrix"),
+           py::arg("row"))
       .def(py::init([](py::buffer b) {
         py::buffer_info info = b.request();
         if (info.format != py::format_descriptor<float>::format()) {
@@ -1242,6 +1245,9 @@ void pybind_kaldi_vector(py::module& m) {
     ));
 
   py::class_<SubVector<double>, VectorBase<double>>(m, "DoubleSubVector")
+      .def(py::init<const MatrixBase<double> &, MatrixIndexT>(),
+           py::arg("matrix"),
+           py::arg("row"))
       .def(py::init([](py::buffer b) {
         py::buffer_info info = b.request();
         if (info.format != py::format_descriptor<double>::format()) {

extensions/util/util.cpp

@@ -710,7 +710,7 @@ void init_util(py::module &_m) {
             py::arg("other"))
         .def("__repr__",
               [](const CtmInterval &a) {
-                return "<CtmInterval of labeled '" + a.label + "' from " + std::to_string(a.begin) + " to " + std::to_string(a.end) +">";
+                return "<CtmInterval of labeled '" + a.label +"("+ std::to_string(a.symbol) +")" + "' from " + std::to_string(a.begin) + " to " + std::to_string(a.end) +" confidence="+std::to_string(a.confidence)+">";
             })
         .def(py::pickle(
             [](const CtmInterval &p) { // __getstate__