Properly display KaldiFatalError messages

.github/workflows/main.yml

@@ -19,7 +19,6 @@ jobs:
         include:
           - os: ubuntu-latest
             label: linux-64
-            prefix: /usr/share/miniconda3/envs/my-env
 
           #- os: macos-latest
           #  label: osx-64
@@ -37,17 +36,15 @@ jobs:
           fetch-depth: 0
 
       - name: Install Conda environment with Micromamba
-        uses: mamba-org/setup-micromamba@v1
+        uses: conda-incubator/setup-miniconda@v3
         with:
           environment-file: environment.yml
-          environment-name: my-env
-          create-args: >-
-            python=3.10
-          cache-environment: true
+          activate-environment: kalpy_build
+          miniforge-version: latest
 
       - name: Build kalpy
         env:
-          KALDI_ROOT: ~/micromamba/envs/my-env
+          KALDI_ROOT: ~/miniconda3/envs/kalpy_build
         shell: bash -l {0}
         run: pip install .
 

extensions/util/util.cpp

@@ -69,10 +69,27 @@ void pybind_arg(py::module& m, Opt& opt) {
           py::arg("name"), py::arg("arg"), py::arg("doc"));
 }
 
+
 }  // namespace
 
 void init_util(py::module &_m) {
   py::module m = _m.def_submodule("util", "util pybind for Kaldi");
+
+
+    PYBIND11_CONSTINIT static py::gil_safe_call_once_and_store<py::object> exc_storage;
+
+    exc_storage.call_once_and_store_result(
+        [&]() { return py::exception<KaldiFatalError>(m, "KaldiFatalError"); });
+
+    py::register_exception_translator([](std::exception_ptr p) {
+        try {
+            if (p) std::rethrow_exception(p);
+        } catch (const KaldiFatalError &e) {
+            py::set_error(exc_storage.get_stored(), e.KaldiMessage());
+        }
+    });
+  //py::register_exception<KaldiFatalError>(m, "PyKaldiFatalError", PyExc_RuntimeError);
+  //py::register_local_exception<KaldiFatalError>(m, "PyKaldiFatalError", PyExc_RuntimeError);
   pybind_basic_vector_holder<int32>(m, "IntVectorHolder");
   py::class_<std::istream>(m, "istream");
   {

kalpy/fstext/lexicon.py

@@ -301,7 +301,7 @@ def compute_disambiguation_symbols(self):
         for pron in self.pronunciations:
 
             phones = pron.pronunciation.split()
-            while phones:
+            while len(phones) > 0:
                 subsequences.add(" ".join(phones))
                 phones = phones[:-1]
         last_used = collections.defaultdict(int)

kalpy/gmm/align.py

@@ -3,6 +3,8 @@
 
 import logging
 import pathlib
+import sys
+import traceback
 import typing
 
 from _kalpy.fstext import VectorFst
@@ -101,15 +103,21 @@ def align_utterances(
             except KeyError:
                 logger.warning(f"Skipping {utterance_id} due to missing training graph")
                 continue
-            alignment = self.align_utterance(training_graph, feats, utterance_id)
-            if alignment is None:
-                yield utterance_id, None
-                num_error += 1
-                continue
-            yield alignment
-            total_likelihood += alignment.likelihood
-            total_frames += len(alignment.alignment)
-            num_done += 1
+            try:
+                alignment = self.align_utterance(training_graph, feats, utterance_id)
+                if alignment is None:
+                    yield utterance_id, None
+                    num_error += 1
+                    continue
+                yield alignment
+                total_likelihood += alignment.likelihood
+                total_frames += len(alignment.alignment)
+                num_done += 1
+            except Exception as e:
+                exc_type, exc_value, exc_traceback = sys.exc_info()
+                logger.warning(f"Error on {utterance_id}: {e}")
+                traceback_lines = traceback.format_exception(exc_type, exc_value, exc_traceback)
+                logger.debug("\n".join(traceback_lines))
         if total_frames:
             logger.info(
                 f"Overall log-likelihood per frame is {total_likelihood / total_frames} over {total_frames} frames."
@@ -153,6 +161,8 @@ def export_alignments(
                     likelihood_writer.Write(
                         str(alignment.utterance_id), alignment.per_frame_likelihoods
                     )
+        except Exception as e:
+            logger.error(e)
         finally:
             writer.Close()
             if word_writer is not None: