diff --git a/test/test_utils.py b/test/test_utils.py
index 870032d6c..3928173c8 100644
--- a/test/test_utils.py
+++ b/test/test_utils.py
@@ -102,6 +102,7 @@ from youtube_dl.utils import (
     url_or_none,
     urljoin,
     urlencode_postdata,
+    urlhandle_detect_ext,
     urshift,
     update_url_query,
     variadic,
@@ -2039,6 +2040,30 @@ Line 1
             'a', 'b', 'c', 'd',
             from_dict={'a': 'c', 'c': [], 'b': 'd', 'd': None}), 'c-d')
 
+    def test_urlhandle_detect_ext(self):
+
+        class UrlHandle(object):
+            _info = {}
+
+            def __init__(self, info):
+                self._info = info
+
+            @property
+            def headers(self):
+                return self._info
+
+        # header with non-ASCII character and contradictory Content-Type
+        urlh = UrlHandle({
+            'Content-Disposition': b'attachment; filename="Epis\xf3dio contains non-ASCI ISO 8859-1 character.mp3"',
+            'Content-Type': b'audio/aac',
+        })
+        self.assertEqual(urlhandle_detect_ext(urlh), 'mp3')
+        # header with no Content-Disposition
+        urlh = UrlHandle({
+            'Content-Type': b'audio/mp3',
+        })
+        self.assertEqual(urlhandle_detect_ext(urlh), 'mp3')
+
 
 if __name__ == '__main__':
     unittest.main()
diff --git a/youtube_dl/utils.py b/youtube_dl/utils.py
index 970bc591a..f35c71601 100644
--- a/youtube_dl/utils.py
+++ b/youtube_dl/utils.py
@@ -4743,7 +4743,8 @@ def parse_codecs(codecs_str):
 
 
 def urlhandle_detect_ext(url_handle):
-    getheader = url_handle.headers.get
+    getheader = lambda x: _decode_compat_str(
+        url_handle.headers.get(x), encoding='iso-8859-1', errors='ignore', or_none=True)
 
     cd = getheader('Content-Disposition')
     if cd:
@@ -4753,6 +4754,11 @@ def urlhandle_detect_ext(url_handle):
             if e:
                 return e
 
+    # from yt-dlp/2647c933b8: thx bashonly
+    meta_ext = (getheader('x-amz-meta-name') or '').rpartition('.')[2]
+    if meta_ext:
+        return meta_ext
+
     return mimetype2ext(getheader('Content-Type'))