diff --git a/npybabel.py b/npybabel.py
index 0c79e5d7d1e..3d9deaec676 100755
--- a/npybabel.py
+++ b/npybabel.py
@@ -4,14 +4,17 @@ __requires__ = 'Babel==0.9.6'
 import sys
 from pkg_resources import load_entry_point
 import re
+import json
 
 if __name__ == '__main__':
     sys.exit(
         load_entry_point('Babel==0.9.6', 'console_scripts', 'pybabel')()
     )
     
-QWEB_EXPR = re.compile(r"""(?:\< *t\-tr *\>(.*?)\< *\/t\-tr *\>)|(?:\_t *\( *((?:\".*?\")|(?:\'.*?\')) *\))""")
-    
+QWEB_EXPR = re.compile(r"""(?:\< *t\-tr *\>(.*?)\< *\/t\-tr *\>)|(?:\_t *\( *((?:"(?:[^"\\]|\\.)*")|(?:'(?:[^'\\]|\\.)*')) *\))""")
+XML_GROUP = 1
+JS_GROUP = 2
+
 def extract_qweb(fileobj, keywords, comment_tags, options):
     """Extract messages from XXX files.
     :param fileobj: the file-like object the messages should be extracted
@@ -31,10 +34,13 @@ def extract_qweb(fileobj, keywords, comment_tags, options):
     index = 0
     line_nbr = 0
     for f in found:
-        group = 1 if f.group(1) else 2
+        group = XML_GROUP if f.group(XML_GROUP) else JS_GROUP
+        mes = f.group(group)
+        if group == JS_GROUP:
+            mes = json.loads(mes)
         while index < f.start():
             if content[index] == "\n":
                 line_nbr += 1
             index += 1
-        result.append((line_nbr, None, f.group(group), ""))
+        result.append((line_nbr, None, mes, ""))
     return result