diff --git a/SpeechMod/Voice/WindowsSpeech.cs b/SpeechMod/Voice/WindowsSpeech.cs
index 4746891..a9767ab 100644
--- a/SpeechMod/Voice/WindowsSpeech.cs
+++ b/SpeechMod/Voice/WindowsSpeech.cs
@@ -78,7 +78,11 @@ private string FormatGenderSpecificVoices(string text)
 
     private void SpeakInternal(string text, float delay = 0f)
     {
-        text = SpeakBegin + text + SpeakEnd;
+        // Send raw SAPI XML (a "<voice .../>" fragment) as-is. Do NOT wrap it in a W3C
+        // <speak xmlns="http://www.w3.org/2001/10/synthesis"> document: that wrapper makes SAPI
+        // ignore the proprietary <voice required="Name=..."> selection (so it falls back to the
+        // system default voice) and makes neural engines such as NaturalVoiceSAPIAdapter reject
+        // the markup with SPERR_UNSUPPORTED_FORMAT (0x80045003) -> no audio.
         if (Main.Settings?.LogVoicedLines == true)
             Debug.Log(text);
         WindowsVoiceUnity.Speak(text, Length(text), delay);