diff --git a/src/calibre/utils/windows/winspeech.cpp b/src/calibre/utils/windows/winspeech.cpp
index d51392d6c4..b75d23bfa8 100644
--- a/src/calibre/utils/windows/winspeech.cpp
+++ b/src/calibre/utils/windows/winspeech.cpp
@@ -205,20 +205,24 @@ public:
                 return serialize_string_for_json(s, out);
             case DT_LIST: {
                 out << '[';
+                bool first = true;
                 for (auto const &i : list) {
+                    if (!first) out << ", ";
+                    first = false;
                     i.serialize(out);
-                    out << ", ";
                 }
                 out << ']';
                 break;
             }
             case DT_OBJECT: {
                 out << '{';
+                bool first = true;
                 for (const auto& [key, value]: object) {
+                    if (!first) out << ", ";
+                    first = false;
                     serialize_string_for_json(key, out);
                     out << ": ";
                     value.serialize(out);
-                    out << ", ";
                 }
                 out << '}';
                 break;
diff --git a/src/calibre/utils/windows/winspeech.py b/src/calibre/utils/windows/winspeech.py
index b2d66ffe86..6a29a217c4 100644
--- a/src/calibre/utils/windows/winspeech.py
+++ b/src/calibre/utils/windows/winspeech.py
@@ -2,22 +2,34 @@
 # License: GPLv3 Copyright: 2023, Kovid Goyal <kovid at kovidgoyal.net>
 
 
+import json
 import sys
 import time
 from contextlib import closing
+from queue import Queue
 from threading import Thread
 
 from calibre.utils.ipc.simple_worker import start_pipe_worker
 
 
+def decode_msg(line: bytes) -> dict:
+    parts = line.strip().split(b' ', 2)
+    msg_id, msg_type, ans = int(parts[0]), parts[1].decode(), json.loads(parts[2])
+    ans['related_to'] = msg_id
+    ans['payload_type'] = msg_type
+    return ans
+
+
 def develop_speech(text='Lucca brazzi sleeps with the fishes'):
     p = start_pipe_worker('from calibre_extensions.winspeech import run_main_loop; run_main_loop()')
-    print('\x1b[32mSpeaking', text, '\x1b[39m', flush=True)
+    print('\x1b[32mSpeaking', text, '\x1b[39m', flush=True)  # ]]
+    q = Queue()
 
     def echo_output(p):
         for line in p.stdout:
-            sys.stdout.buffer.write(b'\x1b[33m' + line + b'\x1b[39m')
+            sys.stdout.buffer.write(b'\x1b[33m' + line + b'\x1b[39m')  # ]]
             sys.stdout.buffer.flush()
+            q.put(decode_msg(line))
 
     def send(*a):
         cmd = ' '.join(map(str, a)) + '\n'
@@ -29,10 +41,13 @@ def develop_speech(text='Lucca brazzi sleeps with the fishes'):
         try:
             send('1 echo Synthesizer started')
             send('2 speak text inline', text)
-            time.sleep(6)
+            while True:
+                m = q.get()
+                if m['payload_type'] == 'media_state_changed' and m['state'] == 'ended':
+                    break
             send('3 echo Synthesizer exiting')
             send('exit')
-            time.sleep(1)
+            p.wait(1)
         finally:
             if p.poll() is None:
                 p.kill()