Merge pull request #692 from gkucsko/c_lev_package

editdistance package for fast WER calculation

nemo/collections/asr/metrics.py

@@ -1,30 +1,13 @@
 # Copyright (c) 2019 NVIDIA Corporation
 from typing import List, Optional
 
+import editdistance
 import torch
 
 
-def __levenshtein(a: List, b: List) -> int:
-    """Calculates the Levenshtein distance between a and b.
-    The code was copied from: http://hetland.org/coding/python/levenshtein.py
-    """
-    n, m = len(a), len(b)
-    if n > m:
-        # Make sure n <= m, to use O(min(n,m)) space
-        a, b = b, a
-        n, m = m, n
-
-    current = list(range(n + 1))
-    for i in range(1, m + 1):
-        previous, current = current, [i] + [0] * n
-        for j in range(1, n + 1):
-            add, delete = previous[j] + 1, current[j - 1] + 1
-            change = previous[j - 1]
-            if a[j - 1] != b[i - 1]:
-                change = change + 1
-            current[j] = min(add, delete, change)
-
-    return current[n]
+def __levenshtein(a: List[str], b: List[str]) -> int:
+    """Calculates the Levenshtein distance between a and b."""
+    return editdistance.eval(a, b)
 
 
 def word_error_rate(hypotheses: List[str], references: List[str], use_cer=False) -> float:

requirements/requirements_asr.txt

@@ -1,4 +1,5 @@
 braceexpand
+editdistance
 frozendict
 inflect
 kaldi-io