diff --git a/proto/backend.proto b/proto/backend.proto
index 2e0f76291..2522639eb 100644
--- a/proto/backend.proto
+++ b/proto/backend.proto
@@ -17,8 +17,7 @@ message BackendInput {
         RenderCardIn render_card = 21;
         int64 local_minutes_west = 22;
         string strip_av_tags = 23;
-        string get_av_tags = 24;
-        string flag_av_tags = 25;
+        ExtractAVTagsIn extract_av_tags = 24;
     }
 }
 
@@ -32,8 +31,7 @@ message BackendOutput {
         RenderCardOut render_card = 21;
         sint32 local_minutes_west = 22;
         string strip_av_tags = 23;
-        GetAVTagsOut get_av_tags = 24;
-        string flag_av_tags = 25;
+        ExtractAVTagsOut extract_av_tags = 24;
 
         BackendError error = 2047;
     }
@@ -149,8 +147,14 @@ message RenderedTemplateReplacement {
     repeated string filters = 3;
 }
 
-message GetAVTagsOut {
-    repeated AVTag av_tags = 1;
+message ExtractAVTagsIn {
+    string text = 1;
+    bool question_side = 2;
+}
+
+message ExtractAVTagsOut {
+    string text = 1;
+    repeated AVTag av_tags = 2;
 }
 
 message AVTag {
diff --git a/pylib/anki/cards.py b/pylib/anki/cards.py
index e8a924b53..10c67d9a9 100644
--- a/pylib/anki/cards.py
+++ b/pylib/anki/cards.py
@@ -5,12 +5,13 @@ from __future__ import annotations
 
 import pprint
 import time
-from typing import Any, Dict, Optional, Union
+from typing import Any, Dict, List, Optional, Union
 
 import anki  # pylint: disable=unused-import
 from anki import hooks
 from anki.consts import *
 from anki.notes import Note
+from anki.sound import AVTag
 from anki.utils import intTime, joinFields, timestampID
 
 # Cards
@@ -26,7 +27,7 @@ from anki.utils import intTime, joinFields, timestampID
 
 
 class Card:
-    _qa: Optional[Dict[str, Union[str, int]]]
+    _qa: Optional[Dict[str, Union[str, int, List[AVTag]]]]
     _note: Optional[Note]
     timerStarted: Optional[float]
     lastIvl: int
@@ -149,6 +150,12 @@ lapses=?, left=?, odue=?, odid=?, did=? where id = ?""",
     def a(self) -> str:
         return self.css() + self._getQA()["a"]
 
+    def question_av_tags(self) -> List[AVTag]:
+        return self._qa["q_av_tags"]  # type: ignore
+
+    def answer_av_tags(self) -> List[AVTag]:
+        return self._qa["a_av_tags"]  # type: ignore
+
     def css(self) -> str:
         return "<style>%s</style>" % self.model()["css"]
 
diff --git a/pylib/anki/collection.py b/pylib/anki/collection.py
index 07069ff99..d87ebe697 100644
--- a/pylib/anki/collection.py
+++ b/pylib/anki/collection.py
@@ -30,8 +30,9 @@ from anki.notes import Note
 from anki.rsbackend import RustBackend
 from anki.sched import Scheduler as V1Scheduler
 from anki.schedv2 import Scheduler as V2Scheduler
+from anki.sound import AVTag
 from anki.tags import TagManager
-from anki.template import QAData, TemplateRenderContext, render_card
+from anki.template import QAData, RenderOutput, TemplateRenderContext, render_card
 from anki.utils import (
     devMode,
     fieldChecksum,
@@ -633,7 +634,7 @@ where c.nid = n.id and c.id in %s group by nid"""
     # data is [cid, nid, mid, did, ord, tags, flds, cardFlags]
     def _renderQA(
         self, data: QAData, qfmt: Optional[str] = None, afmt: Optional[str] = None
-    ) -> Dict[str, Union[str, int]]:
+    ) -> Dict[str, Union[str, int, List[AVTag]]]:
         # extract info from data
         split_fields = splitFields(data[6])
         card_ord = data[4]
@@ -671,31 +672,44 @@ where c.nid = n.id and c.id in %s group by nid"""
         # render fields. if any custom filters are encountered,
         # the field_filter hook will be called.
         try:
-            qtext, atext = render_card(self, qfmt, afmt, ctx)
+            output = render_card(self, qfmt, afmt, ctx)
         except anki.rsbackend.BackendException as e:
             errmsg = _("Card template has a problem:") + f"<br>{e}"
-            qtext = errmsg
-            atext = errmsg
+            output = RenderOutput(
+                question_text=errmsg,
+                answer_text=errmsg,
+                question_av_tags=[],
+                answer_av_tags=[],
+            )
 
         # avoid showing the user a confusing blank card if they've
         # forgotten to add a cloze deletion
         if model["type"] == MODEL_CLOZE:
             if not self.models._availClozeOrds(model, data[6], False):
-                qtext = (
-                    qtext
-                    + "<p>"
-                    + _("Please edit this note and add some cloze deletions. (%s)")
-                    % ("<a href=%s#cloze>%s</a>" % (HELP_SITE, _("help")))
-                )
+                output.question_text += "<p>" + _(
+                    "Please edit this note and add some cloze deletions. (%s)"
+                ) % ("<a href=%s#cloze>%s</a>" % (HELP_SITE, _("help")))
 
         # allow add-ons to modify the generated result
-        (qtext, atext) = hooks.card_did_render((qtext, atext), ctx)
+        (output.question_text, output.answer_text) = hooks.card_did_render(
+            (output.question_text, output.answer_text), ctx
+        )
 
         # legacy hook
-        qtext = runFilter("mungeQA", qtext, "q", fields, model, data, self)
-        atext = runFilter("mungeQA", atext, "a", fields, model, data, self)
+        output.question_text = runFilter(
+            "mungeQA", output.question_text, "q", fields, model, data, self
+        )
+        output.answer_text = runFilter(
+            "mungeQA", output.answer_text, "a", fields, model, data, self
+        )
 
-        return dict(q=qtext, a=atext, id=card_id)
+        return dict(
+            q=output.question_text,
+            a=output.answer_text,
+            id=card_id,
+            q_av_tags=output.question_av_tags,
+            a_av_tags=output.answer_av_tags,
+        )
 
     def _qaData(self, where="") -> Any:
         "Return [cid, nid, mid, did, ord, tags, flds, cardFlags] db query"
diff --git a/pylib/anki/rsbackend.py b/pylib/anki/rsbackend.py
index 0dc726716..00738588b 100644
--- a/pylib/anki/rsbackend.py
+++ b/pylib/anki/rsbackend.py
@@ -161,15 +161,16 @@ class RustBackend:
     def strip_av_tags(self, text: str) -> str:
         return self._run_command(pb.BackendInput(strip_av_tags=text)).strip_av_tags
 
-    def get_av_tags(self, text: str) -> List[AVTag]:
-        return list(
-            map(
-                av_tag_to_native,
-                self._run_command(
-                    pb.BackendInput(get_av_tags=text)
-                ).get_av_tags.av_tags,
+    def extract_av_tags(
+        self, text: str, question_side: bool
+    ) -> Tuple[str, List[AVTag]]:
+        out = self._run_command(
+            pb.BackendInput(
+                extract_av_tags=pb.ExtractAVTagsIn(
+                    text=text, question_side=question_side
+                )
             )
-        )
+        ).extract_av_tags
+        native_tags = list(map(av_tag_to_native, out.av_tags))
 
-    def flag_av_tags(self, text: str) -> str:
-        return self._run_command(pb.BackendInput(flag_av_tags=text)).flag_av_tags
+        return out.text, native_tags
diff --git a/pylib/anki/sound.py b/pylib/anki/sound.py
index fad5b6789..0d5f6d736 100644
--- a/pylib/anki/sound.py
+++ b/pylib/anki/sound.py
@@ -4,8 +4,7 @@
 """
 Sound/TTS references extracted from card text.
 
-Use collection.backend.strip_av_tags(string) to remove all tags,
-and collection.backend.get_av_tags(string) to get a list of AVTags.
+These can be accessed via eg card.question_av_tags()
 """
 
 from __future__ import annotations
diff --git a/pylib/anki/template.py b/pylib/anki/template.py
index 29ad541e2..5604cc87b 100644
--- a/pylib/anki/template.py
+++ b/pylib/anki/template.py
@@ -29,12 +29,14 @@ template_legacy.py file, using the legacy addHook() system.
 from __future__ import annotations
 
 import re
+from dataclasses import dataclass
 from typing import Any, Dict, List, Optional, Tuple
 
 import anki
 from anki import hooks
 from anki.models import NoteType
 from anki.rsbackend import TemplateReplacementList
+from anki.sound import AVTag
 
 QAData = Tuple[
     # Card ID this QA comes from. Corresponds to 'cid' column.
@@ -122,18 +124,34 @@ class TemplateRenderContext:
         return self._note_type
 
 
+@dataclass
+class RenderOutput:
+    question_text: str
+    answer_text: str
+    question_av_tags: List[AVTag]
+    answer_av_tags: List[AVTag]
+
+
 def render_card(
     col: anki.storage._Collection, qfmt: str, afmt: str, ctx: TemplateRenderContext
-) -> Tuple[str, str]:
-    """Renders the provided templates, returning rendered q & a text.
+) -> RenderOutput:
+    """Renders the provided templates, returning rendered output.
 
     Will raise if the template is invalid."""
     (qnodes, anodes) = col.backend.render_card(qfmt, afmt, ctx.fields(), ctx.card_ord())
 
     qtext = apply_custom_filters(qnodes, ctx, front_side=None)
-    atext = apply_custom_filters(anodes, ctx, front_side=qtext)
+    qtext, q_avtags = col.backend.extract_av_tags(qtext, True)
 
-    return qtext, atext
+    atext = apply_custom_filters(anodes, ctx, front_side=qtext)
+    atext, a_avtags = col.backend.extract_av_tags(atext, False)
+
+    return RenderOutput(
+        question_text=qtext,
+        answer_text=atext,
+        question_av_tags=q_avtags,
+        answer_av_tags=a_avtags,
+    )
 
 
 def apply_custom_filters(
@@ -153,7 +171,7 @@ def apply_custom_filters(
         else:
             # do we need to inject in FrontSide?
             if node.field_name == "FrontSide" and front_side is not None:
-                node.current_text = ctx.col().backend.strip_av_tags(front_side)
+                node.current_text = front_side
 
             field_text = node.current_text
             for filter_name in node.filters:
diff --git a/pylib/tests/test_collection.py b/pylib/tests/test_collection.py
index 43ce5d058..829332a3f 100644
--- a/pylib/tests/test_collection.py
+++ b/pylib/tests/test_collection.py
@@ -142,7 +142,7 @@ def test_furigana():
     # and should avoid sound
     n["Front"] = "foo[sound:abc.mp3]"
     n.flush()
-    assert "sound:" in c.q(reload=True)
+    assert "anki:play" in c.q(reload=True)
     # it shouldn't throw an error while people are editing
     m["tmpls"][0]["qfmt"] = "{{kana:}}"
     mm.save(m)
diff --git a/qt/aqt/browser.py b/qt/aqt/browser.py
index 9b9790a2f..e0920cc70 100644
--- a/qt/aqt/browser.py
+++ b/qt/aqt/browser.py
@@ -32,7 +32,6 @@ from aqt.utils import (
     askUser,
     getOnlyText,
     getTag,
-    mungeQA,
     openHelp,
     qtMenuShortcutWorkaround,
     restoreGeom,
@@ -1682,9 +1681,9 @@ where id in %s"""
             # need to force reload even if answer
             txt = c.q(reload=True)
 
-            questionAudio = []
+            question_audio = []
             if self._previewBothSides:
-                questionAudio = self.mw.col.backend.get_av_tags(txt)
+                question_audio = c.question_av_tags()
             if self._previewState == "answer":
                 func = "_showAnswer"
                 txt = c.a()
@@ -1694,13 +1693,13 @@ where id in %s"""
 
             if self.mw.reviewer.autoplay(c):
                 # if we're showing both sides at once, play question audio first
-                av_player.play_tags(questionAudio)
+                av_player.play_tags(question_audio)
                 # then play any audio that hasn't already been played
-                answer_audio = self.mw.col.backend.get_av_tags(txt)
-                unplayed_audio = [x for x in answer_audio if x not in questionAudio]
+                answer_audio = c.answer_av_tags()
+                unplayed_audio = [x for x in answer_audio if x not in question_audio]
                 av_player.extend_and_play(unplayed_audio)
 
-            txt = mungeQA(self.col, txt)
+            txt = self.mw.prepare_card_text_for_display(txt)
             gui_hooks.card_will_show(
                 txt, c, "preview" + self._previewState.capitalize()
             )
diff --git a/qt/aqt/clayout.py b/qt/aqt/clayout.py
index 5ab4bafe2..c518abe91 100644
--- a/qt/aqt/clayout.py
+++ b/qt/aqt/clayout.py
@@ -5,8 +5,10 @@
 import collections
 import json
 import re
+from typing import Optional
 
 import aqt
+from anki.cards import Card
 from anki.consts import *
 from anki.lang import _, ngettext
 from anki.utils import isMac, isWin, joinFields
@@ -18,7 +20,6 @@ from aqt.utils import (
     askUser,
     downArrow,
     getOnlyText,
-    mungeQA,
     openHelp,
     restoreGeom,
     saveGeom,
@@ -29,6 +30,8 @@ from aqt.webview import AnkiWebView
 
 
 class CardLayout(QDialog):
+    card: Optional[Card]
+
     def __init__(self, mw, note, ord=0, parent=None, addMode=False):
         QDialog.__init__(self, parent or mw, Qt.Window)
         mw.setupDialogGC(self)
@@ -339,10 +342,10 @@ Please create a new card type first."""
 
         bodyclass = theme_manager.body_classes_for_card_ord(c.ord)
 
-        q = ti(mungeQA(self.mw.col, c.q(reload=True)))
+        q = ti(self.mw.prepare_card_text_for_display(c.q(reload=True)))
         q = gui_hooks.card_will_show(q, c, "clayoutQuestion")
 
-        a = ti(mungeQA(self.mw.col, c.a()), type="a")
+        a = ti(self.mw.prepare_card_text_for_display(c.a()), type="a")
         a = gui_hooks.card_will_show(a, c, "clayoutAnswer")
 
         # use _showAnswer to avoid the longer delay
@@ -350,7 +353,7 @@ Please create a new card type first."""
         self.pform.backWeb.eval("_showAnswer(%s, '%s');" % (json.dumps(a), bodyclass))
 
         if c.id not in self.playedAudio:
-            av_player.play_from_text(self.mw.col, c.q() + c.a())
+            av_player.play_tags(c.question_av_tags() + c.answer_av_tags())
             self.playedAudio[c.id] = True
 
         self.updateCardNames()
diff --git a/qt/aqt/legacy.py b/qt/aqt/legacy.py
index 50cb959ce..ea5dec1f5 100644
--- a/qt/aqt/legacy.py
+++ b/qt/aqt/legacy.py
@@ -24,11 +24,8 @@ def bodyClass(col, card) -> str:
 
 def allSounds(text) -> List:
     print("allSounds() deprecated")
-    return [
-        x.filename
-        for x in aqt.mw.col.backend.get_av_tags(text)
-        if isinstance(x, SoundOrVideoTag)
-    ]
+    text, tags = aqt.mw.col.backend.extract_av_tags(text, True)
+    return [x.filename for x in tags if isinstance(x, SoundOrVideoTag)]
 
 
 def stripSounds(text) -> str:
diff --git a/qt/aqt/main.py b/qt/aqt/main.py
index 650ed215d..d8e9ff8cb 100644
--- a/qt/aqt/main.py
+++ b/qt/aqt/main.py
@@ -400,14 +400,17 @@ close the profile or restart Anki."""
     def setupSound(self) -> None:
         aqt.sound.setup_audio(self.taskman, self.pm.base)
 
-    def process_av_tags(self, text: str) -> Tuple[str, List[AVTag]]:
-        "Return card text with play buttons added, and the extracted AV tags."
-        tags = self.col.backend.get_av_tags(text)
+    def _add_play_buttons(self, text: str) -> str:
+        "Return card text with play buttons added, or stripped."
         if self.pm.profile.get("showPlayButtons", True):
-            text = aqt.sound.av_flags_to_html(self.col.backend.flag_av_tags(text))
+            return aqt.sound.av_refs_to_play_icons(text)
         else:
-            text = self.col.backend.strip_av_tags(text)
-        return (text, tags)
+            return aqt.sound.strip_av_refs(text)
+
+    def prepare_card_text_for_display(self, text: str) -> str:
+        text = self.col.media.escapeImages(text)
+        text = self._add_play_buttons(text)
+        return text
 
     # Collection load/unload
     ##########################################################################
diff --git a/qt/aqt/reviewer.py b/qt/aqt/reviewer.py
index c7c7371c5..163f6dd89 100644
--- a/qt/aqt/reviewer.py
+++ b/qt/aqt/reviewer.py
@@ -13,20 +13,13 @@ from typing import List, Optional
 from anki import hooks
 from anki.cards import Card
 from anki.lang import _, ngettext
-from anki.sound import AVTag
 from anki.utils import stripHTML
 from aqt import AnkiQt, gui_hooks
 from aqt.qt import *
-from aqt.sound import av_player, getAudio
+from aqt.sound import av_player, getAudio, play_clicked_audio
 from aqt.theme import theme_manager
 from aqt.toolbar import BottomBar
-from aqt.utils import (
-    askUserDialog,
-    downArrow,
-    mungeQA,
-    qtMenuShortcutWorkaround,
-    tooltip,
-)
+from aqt.utils import askUserDialog, downArrow, qtMenuShortcutWorkaround, tooltip
 
 
 class Reviewer:
@@ -40,7 +33,6 @@ class Reviewer:
         self.hadCardQueue = False
         self._answeredIds: List[int] = []
         self._recordedAudio = None
-        self._current_side_audio: Optional[List[AVTag]] = None
         self.typeCorrect = None  # web init happens before this is set
         self.state: Optional[str] = None
         self.bottom = BottomBar(mw, mw.bottomWeb)
@@ -116,21 +108,12 @@ class Reviewer:
             state = self.state
             c = self.card
         if state == "question":
-            av_player.play_from_text(self.mw.col, c.q())
+            av_player.play_tags(c.question_av_tags())
         elif state == "answer":
-            txt = ""
+            tags = c.answer_av_tags()
             if self._replayq(c, previewer):
-                txt = c.q()
-            txt += c.a()
-            av_player.play_from_text(self.mw.col, txt)
-
-    def on_play_button(self, idx: int):
-        try:
-            tag = self._current_side_audio[idx]
-        except IndexError:
-            return
-
-        av_player.play_tags([tag])
+                tags = c.question_av_tags() + tags
+            av_player.play_tags(tags)
 
     # Initializing the webview
     ##########################################################################
@@ -176,7 +159,7 @@ class Reviewer:
     ##########################################################################
 
     def _mungeQA(self, buf):
-        return self.typeAnsFilter(mungeQA(self.mw.col, buf))
+        return self.typeAnsFilter(self.mw.prepare_card_text_for_display(buf))
 
     def _showQuestion(self) -> None:
         self._reps += 1
@@ -192,9 +175,9 @@ The front of this card is empty. Please run Tools>Empty Cards."""
         else:
             q = c.q()
 
-        q, self._current_side_audio = self.mw.process_av_tags(q)
+        # play audio?
         if self.autoplay(c):
-            av_player.play_tags(self._current_side_audio)
+            av_player.play_tags(c.question_av_tags())
 
         # render & update bottom
         q = self._mungeQA(q)
@@ -236,9 +219,9 @@ The front of this card is empty. Please run Tools>Empty Cards."""
         c = self.card
         a = c.a()
         # play audio?
-        a, self._current_side_audio = self.mw.process_av_tags(a)
         if self.autoplay(c):
-            av_player.play_tags(self._current_side_audio)
+            av_player.play_tags(c.answer_av_tags())
+
         a = self._mungeQA(a)
         a = gui_hooks.card_will_show(a, c, "reviewAnswer")
         # render and update bottom
@@ -333,7 +316,7 @@ The front of this card is empty. Please run Tools>Empty Cards."""
         elif url == "more":
             self.showContextMenu()
         elif url.startswith("play:"):
-            self.on_play_button(int(url.split(":")[1]))
+            play_clicked_audio(url, self.card)
         else:
             print("unrecognized anki link:", url)
 
diff --git a/qt/aqt/sound.py b/qt/aqt/sound.py
index 9ab2dabc6..0a7d02e11 100644
--- a/qt/aqt/sound.py
+++ b/qt/aqt/sound.py
@@ -16,8 +16,8 @@ from typing import Any, Callable, Dict, List, Optional, Tuple
 
 import pyaudio
 
-import anki
 import aqt
+from anki.cards import Card
 from anki.lang import _
 from anki.sound import AVTag, SoundOrVideoTag
 from anki.utils import isLin, isMac, isWin
@@ -104,14 +104,6 @@ class AVPlayer:
         self._enqueued.extend(tags)
         self._play_next_if_idle()
 
-    def play_from_text(self, col: anki.storage._Collection, text: str) -> None:
-        tags = col.backend.get_av_tags(text)
-        self.play_tags(tags)
-
-    def extend_from_text(self, col: anki.storage._Collection, text: str) -> None:
-        tags = col.backend.get_av_tags(text)
-        self.extend_and_play(tags)
-
     def stop_and_clear_queue(self) -> None:
         self._enqueued = []
         self._stop_if_playing()
@@ -572,9 +564,7 @@ def play(filename: str) -> None:
 
 
 def playFromText(text) -> None:
-    from aqt import mw
-
-    av_player.extend_from_text(mw.col, text)
+    print("playFromText() deprecated")
 
 
 # legacy globals
@@ -590,19 +580,39 @@ for (k, v) in _exports:
 # Tag handling
 ##########################################################################
 
-AV_FLAG_RE = re.compile(r"\[anki:play\](\d+)\[/anki:play]")
+AV_FLAG_RE = re.compile(r"\[anki:(play:.:\d+)\]")
 
 
-def av_flags_to_html(text: str) -> str:
+def strip_av_refs(text: str) -> str:
+    return AV_FLAG_RE.sub("", text)
+
+
+def av_refs_to_play_icons(text: str) -> str:
+    """Add play icons into the HTML.
+
+    When clicked, the icon will call eg pycmd('play:q:1').
+    """
+
     def repl(match: re.Match) -> str:
         return f"""
-<a class=soundLink href=# onclick="pycmd('play:{match.group(1)}'); return false;">
+<a class=soundLink href=# onclick="pycmd('{match.group(1)}'); return false;">
   <img class=playImage src='/_anki/imgs/play.png'>
 </a>"""
 
     return AV_FLAG_RE.sub(repl, text)
 
 
+def play_clicked_audio(pycmd: str, card: Card) -> None:
+    """eg. if pycmd is 'play:q:0', play the first audio on the question side."""
+    play, context, str_idx = pycmd.split(":")
+    idx = int(str_idx)
+    if context == "q":
+        tags = card.question_av_tags()
+    else:
+        tags = card.answer_av_tags()
+    av_player.play_tags([tags[idx]])
+
+
 # Init defaults
 ##########################################################################
 
diff --git a/qt/aqt/utils.py b/qt/aqt/utils.py
index 2691b4b4f..9143636cd 100644
--- a/qt/aqt/utils.py
+++ b/qt/aqt/utils.py
@@ -480,8 +480,8 @@ def restoreHeader(widget, key):
 
 
 def mungeQA(col, txt):
+    print("mungeQA() deprecated; use mw.prepare_card_text_for_display()")
     txt = col.media.escapeImages(txt)
-    txt = col.backend.strip_av_tags(txt)
     return txt
 
 
diff --git a/rslib/src/backend.rs b/rslib/src/backend.rs
index 201fbe7ce..2df96bd9f 100644
--- a/rslib/src/backend.rs
+++ b/rslib/src/backend.rs
@@ -10,7 +10,7 @@ use crate::template::{
     render_card, without_legacy_template_directives, FieldMap, FieldRequirements, ParsedTemplate,
     RenderedNode,
 };
-use crate::text::{av_tags_in_string, flag_av_tags, strip_av_tags, AVTag};
+use crate::text::{extract_av_tags, strip_av_tags, AVTag};
 use prost::Message;
 use std::collections::{HashMap, HashSet};
 use std::path::PathBuf;
@@ -100,8 +100,7 @@ impl Backend {
                 OValue::LocalMinutesWest(local_minutes_west_for_stamp(stamp))
             }
             Value::StripAvTags(text) => OValue::StripAvTags(strip_av_tags(&text).into()),
-            Value::GetAvTags(text) => OValue::GetAvTags(self.get_av_tags(&text)),
-            Value::FlagAvTags(text) => OValue::FlagAvTags(flag_av_tags(&text).into()),
+            Value::ExtractAvTags(input) => OValue::ExtractAvTags(self.extract_av_tags(input)),
         })
     }
 
@@ -183,11 +182,13 @@ impl Backend {
         })
     }
 
-    fn get_av_tags(&self, text: &str) -> pt::GetAvTagsOut {
-        let tags = av_tags_in_string(text)
+    fn extract_av_tags(&self, input: pt::ExtractAvTagsIn) -> pt::ExtractAvTagsOut {
+        let (text, tags) = extract_av_tags(&input.text, input.question_side);
+        let pt_tags = tags
+            .into_iter()
             .map(|avtag| match avtag {
                 AVTag::SoundOrVideo(file) => pt::AvTag {
-                    value: Some(pt::av_tag::Value::SoundOrVideo(file.to_string())),
+                    value: Some(pt::av_tag::Value::SoundOrVideo(file)),
                 },
                 AVTag::TextToSpeech {
                     field_text,
@@ -196,16 +197,19 @@ impl Backend {
                     other_args,
                 } => pt::AvTag {
                     value: Some(pt::av_tag::Value::Tts(pt::TtsTag {
-                        field_text: field_text.to_string(),
-                        lang: lang.to_string(),
-                        voices: voices.into_iter().map(ToOwned::to_owned).collect(),
-                        other_args: other_args.into_iter().map(ToOwned::to_owned).collect(),
+                        field_text,
+                        lang,
+                        voices,
+                        other_args,
                     })),
                 },
             })
             .collect();
 
-        pt::GetAvTagsOut { av_tags: tags }
+        pt::ExtractAvTagsOut {
+            text: text.into(),
+            av_tags: pt_tags,
+        }
     }
 }
 
diff --git a/rslib/src/template.rs b/rslib/src/template.rs
index 5197fb746..42deccb1e 100644
--- a/rslib/src/template.rs
+++ b/rslib/src/template.rs
@@ -3,7 +3,6 @@
 
 use crate::err::{Result, TemplateError};
 use crate::template_filters::apply_filters;
-use crate::text::strip_av_tags;
 use lazy_static::lazy_static;
 use nom;
 use nom::branch::alt;
@@ -269,7 +268,6 @@ pub(crate) struct RenderContext<'a> {
     pub nonempty_fields: &'a HashSet<&'a str>,
     pub question_side: bool,
     pub card_ord: u16,
-    pub front_text: Option<Cow<'a, str>>,
 }
 
 impl ParsedTemplate<'_> {
@@ -302,19 +300,13 @@ fn render_into(
                 key: key @ "FrontSide",
                 ..
             } => {
-                if let Some(front_side) = &context.front_text {
-                    // a fully rendered front side is available, so we can
-                    // bake it into the output
-                    append_str_to_nodes(rendered_nodes, front_side.as_ref());
-                } else {
-                    // the front side contains unknown filters, and must
-                    // be completed by the Python code
-                    rendered_nodes.push(RenderedNode::Replacement {
-                        field_name: (*key).to_string(),
-                        filters: vec![],
-                        current_text: "".into(),
-                    });
-                }
+                // defer FrontSide rendering to Python, as extra
+                // filters may be required
+                rendered_nodes.push(RenderedNode::Replacement {
+                    field_name: (*key).to_string(),
+                    filters: vec![],
+                    current_text: "".into(),
+                });
             }
             Replacement { key: "", filters } if !filters.is_empty() => {
                 // if a filter is provided, we accept an empty field name to
@@ -435,19 +427,12 @@ pub fn render_card(
         nonempty_fields: &nonempty_fields(field_map),
         question_side: true,
         card_ord,
-        front_text: None,
     };
 
     // question side
     let qnorm = without_legacy_template_directives(qfmt);
     let qnodes = ParsedTemplate::from_text(qnorm.as_ref())?.render(&context)?;
 
-    // if the question side didn't have any unknown filters, we can pass
-    // FrontSide in now
-    if let [RenderedNode::Text { ref text }] = *qnodes.as_slice() {
-        context.front_text = Some(strip_av_tags(text));
-    }
-
     // answer side
     context.question_side = false;
     let anorm = without_legacy_template_directives(afmt);
@@ -517,10 +502,9 @@ mod test {
     use super::{FieldMap, ParsedNode::*, ParsedTemplate as PT};
     use crate::err::TemplateError;
     use crate::template::{
-        field_is_empty, nonempty_fields, render_card, without_legacy_template_directives,
-        FieldRequirements, RenderContext, RenderedNode,
+        field_is_empty, nonempty_fields, without_legacy_template_directives, FieldRequirements,
+        RenderContext,
     };
-    use crate::text::strip_html;
     use std::collections::{HashMap, HashSet};
     use std::iter::FromIterator;
 
@@ -683,7 +667,6 @@ mod test {
             nonempty_fields: &nonempty_fields(&map),
             question_side: true,
             card_ord: 1,
-            front_text: None,
         };
 
         use crate::template::RenderedNode as FN;
@@ -780,39 +763,4 @@ mod test {
             }]
         );
     }
-
-    fn get_complete_template(nodes: &Vec<RenderedNode>) -> Option<&str> {
-        if let [RenderedNode::Text { ref text }] = nodes.as_slice() {
-            Some(text.as_str())
-        } else {
-            None
-        }
-    }
-
-    #[test]
-    fn test_render_full() {
-        // make sure front and back side renders cloze differently
-        let fmt = "{{cloze:Text}}";
-        let clozed_text = "{{c1::one}} {{c2::two::hint}}";
-        let map: HashMap<_, _> = vec![("Text", clozed_text)].into_iter().collect();
-
-        let (qnodes, anodes) = render_card(fmt, fmt, &map, 0).unwrap();
-        assert_eq!(
-            strip_html(get_complete_template(&qnodes).unwrap()),
-            "[...] two"
-        );
-        assert_eq!(
-            strip_html(get_complete_template(&anodes).unwrap()),
-            "one two"
-        );
-
-        // FrontSide should render if only standard modifiers were used
-        let (_qnodes, anodes) =
-            render_card("{{kana:text:Text}}", "{{FrontSide}}", &map, 1).unwrap();
-        assert_eq!(get_complete_template(&anodes).unwrap(), clozed_text);
-
-        // But if a custom modifier was used, it's deferred to the Python code
-        let (_qnodes, anodes) = render_card("{{custom:Text}}", "{{FrontSide}}", &map, 1).unwrap();
-        assert_eq!(get_complete_template(&anodes).is_none(), true)
-    }
 }
diff --git a/rslib/src/template_filters.rs b/rslib/src/template_filters.rs
index b6d4bdae1..4519a480e 100644
--- a/rslib/src/template_filters.rs
+++ b/rslib/src/template_filters.rs
@@ -341,7 +341,6 @@ field</a>
             nonempty_fields: &Default::default(),
             question_side: false,
             card_ord: 0,
-            front_text: None,
         };
         assert_eq!(
             apply_filters("ignored", &["cloze", "type"], "Text", &ctx),
@@ -357,7 +356,6 @@ field</a>
             nonempty_fields: &Default::default(),
             question_side: true,
             card_ord: 0,
-            front_text: None,
         };
         assert_eq!(strip_html(&cloze_filter(text, &ctx)).as_ref(), "[...] two");
         assert_eq!(
diff --git a/rslib/src/text.rs b/rslib/src/text.rs
index 105643bba..154638c25 100644
--- a/rslib/src/text.rs
+++ b/rslib/src/text.rs
@@ -9,13 +9,13 @@ use std::collections::HashSet;
 use std::ptr;
 
 #[derive(Debug, PartialEq)]
-pub enum AVTag<'a> {
-    SoundOrVideo(Cow<'a, str>),
+pub enum AVTag {
+    SoundOrVideo(String),
     TextToSpeech {
-        field_text: Cow<'a, str>,
-        lang: &'a str,
-        voices: Vec<&'a str>,
-        other_args: Vec<&'a str>,
+        field_text: String,
+        lang: String,
+        voices: Vec<String>,
+        other_args: Vec<String>,
     },
 }
 
@@ -78,27 +78,29 @@ pub fn strip_av_tags(text: &str) -> Cow<str> {
     AV_TAGS.replace_all(text, "")
 }
 
-pub fn flag_av_tags(text: &str) -> Cow<str> {
-    let mut idx = 0;
-    AV_TAGS.replace_all(text, |_caps: &Captures| {
-        let text = format!("[anki:play]{}[/anki:play]", idx);
-        idx += 1;
-        text
-    })
-}
-pub fn av_tags_in_string(text: &str) -> impl Iterator<Item = AVTag> {
-    AV_TAGS.captures_iter(text).map(|caps| {
-        if let Some(av_file) = caps.get(1) {
-            AVTag::SoundOrVideo(decode_entities(av_file.as_str()))
+/// Extract audio tags from string, replacing them with [anki:play] refs
+pub fn extract_av_tags<'a>(text: &'a str, question_side: bool) -> (Cow<'a, str>, Vec<AVTag>) {
+    let mut tags = vec![];
+    let context = if question_side { 'q' } else { 'a' };
+    let replaced_text = AV_TAGS.replace_all(text, |caps: &Captures| {
+        // extract
+        let tag = if let Some(av_file) = caps.get(1) {
+            AVTag::SoundOrVideo(decode_entities(av_file.as_str()).into())
         } else {
             let args = caps.get(2).unwrap();
             let field_text = caps.get(3).unwrap();
             tts_tag_from_string(field_text.as_str(), args.as_str())
-        }
-    })
+        };
+        tags.push(tag);
+
+        // and replace with reference
+        format!("[anki:play:{}:{}]", context, tags.len() - 1)
+    });
+
+    (replaced_text, tags)
 }
 
-fn tts_tag_from_string<'a>(field_text: &'a str, args: &'a str) -> AVTag<'a> {
+fn tts_tag_from_string<'a>(field_text: &'a str, args: &'a str) -> AVTag {
     let mut other_args = vec![];
     let mut split_args = args.split(' ');
     let lang = split_args.next().unwrap_or("");
@@ -109,15 +111,15 @@ fn tts_tag_from_string<'a>(field_text: &'a str, args: &'a str) -> AVTag<'a> {
             voices = remaining_arg
                 .split('=')
                 .nth(1)
-                .map(|voices| voices.split(',').collect());
+                .map(|voices| voices.split(',').map(ToOwned::to_owned).collect());
         } else {
-            other_args.push(remaining_arg);
+            other_args.push(remaining_arg.to_owned());
         }
     }
 
     AVTag::TextToSpeech {
-        field_text: strip_html_for_tts(field_text),
-        lang,
+        field_text: strip_html_for_tts(field_text).into(),
+        lang: lang.into(),
         voices: voices.unwrap_or_else(Vec::new),
         other_args,
     }
@@ -149,7 +151,7 @@ pub fn cloze_numbers_in_string(html: &str) -> HashSet<u16> {
 #[cfg(test)]
 mod test {
     use crate::text::{
-        av_tags_in_string, cloze_numbers_in_string, flag_av_tags, strip_av_tags, strip_html,
+        cloze_numbers_in_string, extract_av_tags, strip_av_tags, strip_html,
         strip_html_preserving_image_filenames, AVTag,
     };
     use std::collections::HashSet;
@@ -188,22 +190,21 @@ mod test {
         let s =
             "abc[sound:fo&amp;o.mp3]def[anki:tts][en_US voices=Bob,Jane]foo<br>1&gt;2[/anki:tts]gh";
         assert_eq!(strip_av_tags(s), "abcdefgh");
+
+        let (text, tags) = extract_av_tags(s, true);
+        assert_eq!(text, "abc[anki:play:q:0]def[anki:play:q:1]gh");
+
         assert_eq!(
-            av_tags_in_string(s).collect::<Vec<_>>(),
+            tags,
             vec![
                 AVTag::SoundOrVideo("fo&o.mp3".into()),
                 AVTag::TextToSpeech {
                     field_text: "foo 1>2".into(),
-                    lang: "en_US",
-                    voices: vec!["Bob", "Jane"],
+                    lang: "en_US".into(),
+                    voices: vec!["Bob".into(), "Jane".into()],
                     other_args: vec![]
                 },
             ]
         );
-
-        assert_eq!(
-            flag_av_tags(s),
-            "abc[anki:play]0[/anki:play]def[anki:play]1[/anki:play]gh"
-        );
     }
 }