small bugfix and add asr evaluator to doc (#6229)

* small fix eval_metrics Signed-off-by: fayejf <[email protected]> * fix rng Signed-off-by: fayejf <[email protected]> * add asr evaluator to doc Signed-off-by: fayejf <[email protected]> --------- Signed-off-by: fayejf <[email protected]>
NVIDIA · Mar 17, 2023 · 3bb0002 · 3bb0002
1 parent 7865c52
commit 3bb0002
Show file tree

Hide file tree

Showing 4 changed files with 10 additions and 3 deletions.
diff --git a/docs/source/tools/asr_evaluator.rst b/docs/source/tools/asr_evaluator.rst
@@ -0,0 +1,6 @@
+ASR Evaluator
+========================
+
+ASR evaluator is a tool for thoroughly evaluating the performance of ASR models and other features such as Voice Activity Detection.
+
+See more details in: https://github.com/NVIDIA/NeMo/tree/stable/tools/asr_evaluator
diff --git a/docs/source/tools/intro.rst b/docs/source/tools/intro.rst
@@ -10,6 +10,7 @@ NeMo provides a set of tools useful for developing Automatic Speech Recognitions
    ctc_segmentation
    speech_data_explorer
    comparison_tool
+   asr_evaluator
 
 
 There are also additional NeMo-related tools hosted in separate github repositories:

diff --git a/nemo/collections/asr/parts/preprocessing/perturb.py b/nemo/collections/asr/parts/preprocessing/perturb.py
@@ -362,8 +362,8 @@ def __init__(
             self._audiodataset = AugmentationDataset(manifest_path, audio_tar_filepaths, shuffle_n)
             self._data_iterator = iter(self._audiodataset)
 
-        self.rng = rng
-        random.seed(self.rng) if rng else None
+        self._rng = rng
+        random.seed(self._rng) if rng else None
 
     def perturb(self, data):
         impulse = read_one_audiosegment(

diff --git a/tools/asr_evaluator/utils.py b/tools/asr_evaluator/utils.py
@@ -407,7 +407,7 @@ def cal_target_metadata_wer(manifest: str, target: str, meta_cfg: DictConfig, ev
                     raise ValueError("Current only support target metadata belongs to numeric or string ")
 
         for slot_key in slot_wer:
-            slot_wer[slot_key]['wer'] = slot_wer[slot_key]['errors'] / slot_wer[slot_key]['tokens']
+            slot_wer[slot_key][eval_metric] = slot_wer[slot_key]['errors'] / slot_wer[slot_key]['tokens']
             slot_wer[slot_key]['ins_rate'] = slot_wer[slot_key]['inss'] / slot_wer[slot_key]['tokens']
             slot_wer[slot_key]['del_rate'] = slot_wer[slot_key]['dels'] / slot_wer[slot_key]['tokens']
             slot_wer[slot_key]['sub_rate'] = slot_wer[slot_key]['subs'] / slot_wer[slot_key]['tokens']