diff --git a/mopidy/audio/actor.py b/mopidy/audio/actor.py
index c3d11d98..6aecbe8a 100644
--- a/mopidy/audio/actor.py
+++ b/mopidy/audio/actor.py
@@ -47,6 +47,7 @@ class Audio(pykka.ThreadingActor):
         self._volume_set = None
 
         self._appsrc = None
+        self._appsrc_caps = None
         self._appsrc_need_data_callback = None
         self._appsrc_need_data_id = None
         self._appsrc_enough_data_callback = None
@@ -88,6 +89,7 @@ class Audio(pykka.ThreadingActor):
         source, self._appsrc = self._appsrc, None
         if source is None:
             return
+        self._appsrc_caps = None
         if self._appsrc_need_data_id is not None:
             source.disconnect(self._appsrc_need_data_id)
             self._appsrc_need_data_id = None
@@ -103,13 +105,8 @@ class Audio(pykka.ThreadingActor):
         if not uri or not uri.startswith('appsrc://'):
             return
 
-        # These caps matches the audio data provided by libspotify
-        default_caps = gst.Caps(
-            b'audio/x-raw-int, endianness=(int)1234, channels=(int)2, '
-            b'width=(int)16, depth=(int)16, signed=(boolean)true, '
-            b'rate=(int)44100')
         source = element.get_property('source')
-        source.set_property('caps', default_caps)
+        source.set_property('caps', self._appsrc_caps)
         source.set_property('format', b'time')
         source.set_property('stream-type', b'seekable')
         source.set_property('max-bytes', 1024 * 1024)  # 1 MB
@@ -290,12 +287,16 @@ class Audio(pykka.ThreadingActor):
         """
         self._playbin.set_property('uri', uri)
 
-    def set_appsrc(self, need_data=None, enough_data=None, seek_data=None):
+    def set_appsrc(
+            self, caps, need_data=None, enough_data=None, seek_data=None):
         """
         Switch to using appsrc for getting audio to be played.
 
         You *MUST* call :meth:`prepare_change` before calling this method.
 
+        :param caps: GStreamer caps string describing the audio format to
+            expect
+        :type caps: string
         :param need_data: callback for when appsrc needs data
         :type need_data: callable which takes data length hint in ms
         :param enough_data: callback for when appsrc has enough data
@@ -304,6 +305,9 @@ class Audio(pykka.ThreadingActor):
             to continue playback
         :type seek_data: callable which takes time position in ms
         """
+        if isinstance(caps, unicode):
+            caps = caps.encode('utf-8')
+        self._appsrc_caps = gst.Caps(caps)
         self._appsrc_need_data_callback = need_data
         self._appsrc_enough_data_callback = enough_data
         self._appsrc_seek_data_callback = seek_data
diff --git a/mopidy/backends/spotify/playback.py b/mopidy/backends/spotify/playback.py
index 9855e124..6899ee47 100644
--- a/mopidy/backends/spotify/playback.py
+++ b/mopidy/backends/spotify/playback.py
@@ -27,6 +27,12 @@ def seek_data_callback(spotify_backend, time_position):
 
 
 class SpotifyPlaybackProvider(base.BasePlaybackProvider):
+    # These GStreamer caps matches the audio data provided by libspotify
+    _caps = (
+        'audio/x-raw-int, endianness=(int)1234, channels=(int)2, '
+        'width=(int)16, depth=(int)16, signed=(boolean)true, '
+        'rate=(int)44100')
+
     def play(self, track):
         if track.uri is None:
             return False
@@ -42,6 +48,7 @@ class SpotifyPlaybackProvider(base.BasePlaybackProvider):
 
             self.audio.prepare_change()
             self.audio.set_appsrc(
+                self._caps,
                 need_data=None,
                 enough_data=None,
                 seek_data=seek_data_callback_bound)