diff --git a/conserver/links/wtf_transcribe/__init__.py b/conserver/links/wtf_transcribe/__init__.py
index 7323a35..685e83a 100644
--- a/conserver/links/wtf_transcribe/__init__.py
+++ b/conserver/links/wtf_transcribe/__init__.py
@@ -25,6 +25,7 @@
 
 import json
 import logging
+import mimetypes
 import requests
 from typing import Optional, Dict, Any
 
@@ -180,8 +181,33 @@ def dialog_to_audio_binary(dialog, url_timeout):
 def dialog_filename(dialog, dialog_index):
     return dialog.get("filename", f"audio_{dialog_index}.wav")
 
-def dialog_mimetype(dialog):
-    return dialog.get("mimetype", "audio/wav")
+def dialog_mediatype(dialog):
+    """Resolve a dialog's media type.
+
+    Prefers the modern ``mediatype`` field, then falls back to the legacy
+    ``mimetype`` field (in case a caller passes a pre-compat dialog), then
+    derives it from the dialog's filename extension, and finally defaults
+    to audio/wav.
+    """
+    mediatype = dialog.get("mediatype")
+    if mediatype:
+        return mediatype
+
+    mimetype = dialog.get("mimetype")
+    if mimetype:
+        return mimetype
+
+    filename = dialog.get("filename")
+    if filename:
+        guessed_type, _ = mimetypes.guess_type(filename)
+        if guessed_type and (guessed_type.startswith("audio/") or guessed_type.startswith("video/")):
+            return guessed_type
+
+    return "audio/wav"
+
+
+# Backwards-compatible alias; prefer dialog_mediatype for new code.
+dialog_mimetype = dialog_mediatype
 
 def build_vfun_headers(api_key):
     headers = {}
@@ -202,8 +228,8 @@ def maybe_decode_double_encoded_json(response_json):
 
 def send_audio_to_vfun(audio_binary, dialog, dialog_index, vfun_server_url, api_key, diarize, language, vfun_timeout):
     filename = dialog_filename(dialog, dialog_index)
-    mimetype = dialog_mimetype(dialog)
-    files = {"file-binary": (filename, audio_binary, mimetype)}
+    mediatype = dialog_mediatype(dialog)
+    files = {"file-binary": (filename, audio_binary, mediatype)}
     headers = build_vfun_headers(api_key)
     data = build_vfun_data(diarize, language)
     response = requests.post(
diff --git a/conserver/links/wtf_transcribe/test_wtf_transcribe_unit.py b/conserver/links/wtf_transcribe/test_wtf_transcribe_unit.py
index 85e3e02..2102ca6 100644
--- a/conserver/links/wtf_transcribe/test_wtf_transcribe_unit.py
+++ b/conserver/links/wtf_transcribe/test_wtf_transcribe_unit.py
@@ -10,7 +10,7 @@
     analysis_dialog_index,
     analysis_is_wtf_transcription,
     dialog_filename,
-    dialog_mimetype,
+    dialog_mediatype,
     dialog_to_audio_binary,
     dialog_to_binary,
     dialog_to_index,
@@ -63,7 +63,27 @@ def test_analysis_and_dialog_helpers_cover_transcribed_and_recording_checks():
 def test_file_and_remote_loaders_handle_success_and_failure():
     assert is_file_url("file:///tmp/audio.wav") is True
     assert dialog_filename({}, 3) == "audio_3.wav"
-    assert dialog_mimetype({}) == "audio/wav"
+    assert dialog_mediatype({}) == "audio/wav"
+
+
+def test_dialog_mediatype_resolution_order():
+    # Modern adapters write "mediatype"; it takes priority.
+    assert dialog_mediatype({"mediatype": "audio/mpeg"}) == "audio/mpeg"
+
+    # Legacy "mimetype" is honoured when "mediatype" is absent.
+    assert dialog_mediatype({"mimetype": "audio/mpeg"}) == "audio/mpeg"
+
+    # "mediatype" wins over a legacy "mimetype" when both are present.
+    assert dialog_mediatype({"mediatype": "audio/mpeg", "mimetype": "audio/wav"}) == "audio/mpeg"
+
+    # Neither field present: derive from the filename extension.
+    assert dialog_mediatype({"filename": "clip.ogg"}) == "audio/ogg"
+
+    # A non-audio/video guess from the filename doesn't count; fall through to default.
+    assert dialog_mediatype({"filename": "clip.txt"}) == "audio/wav"
+
+    # Nothing at all gives the audio/wav default.
+    assert dialog_mediatype({}) == "audio/wav"
 
     with patch("builtins.open", mock_open(read_data=b"file-bytes")):
         assert maybe_load_file_url("file:///tmp/audio.wav") == b"file-bytes"