diff --git a/fluster/utils.py b/fluster/utils.py
index 44b9206f..5d7b5d63 100644
--- a/fluster/utils.py
+++ b/fluster/utils.py
@@ -21,6 +21,7 @@
 import contextlib
 import hashlib
 import http.client
+import io
 import os
 import platform
 import random
@@ -320,11 +321,35 @@ def normalize_path(path: str) -> str:
 
 
 def _read_wav(path: str) -> Tuple[array.array[int], int, int]:
-    """Load a WAV file and return (interleaved_samples, n_channels, sampwidth). Supports 16 and 32-bit PCM."""
-    with wave.open(path, "rb") as w:
-        n_channels = w.getnchannels()
-        sampwidth = w.getsampwidth()
-        raw = w.readframes(w.getnframes())
+    """Load a WAV file and return (interleaved_samples, n_channels, sampwidth).
+    Supports 16 and 32-bit PCM and WAVE_FORMAT_EXTENSIBLE (0xFFFE).
+    Python < 3.12 rejects WAVE_FORMAT_EXTENSIBLE, so on failure we locate
+    the fmt chunk and patch only the wFormatTag — the PCM data layout is identical."""
+    try:
+        with wave.open(path, "rb") as w:
+            return _extract_wav(w)
+    except wave.Error:
+        pass
+    # Locate the fmt chunk in the RIFF structure and patch only wFormatTag
+    with open(path, "rb") as f:
+        data = bytearray(f.read())
+    pos = 12  # skip RIFF + size + WAVE
+    while pos < len(data) - 8:
+        chunk_id = data[pos : pos + 4]
+        chunk_size = int.from_bytes(data[pos + 4 : pos + 8], "little")
+        if chunk_id == b"fmt ":
+            if int.from_bytes(data[pos + 8 : pos + 10], "little") == 0xFFFE:
+                data[pos + 8 : pos + 10] = b"\x01\x00"
+            break
+        pos += 8 + chunk_size + (chunk_size % 2)
+    with wave.open(io.BytesIO(bytes(data))) as w:
+        return _extract_wav(w)
+
+
+def _extract_wav(w: wave.Wave_read) -> Tuple[array.array[int], int, int]:
+    n_channels = w.getnchannels()
+    sampwidth = w.getsampwidth()
+    raw = w.readframes(w.getnframes())
     if sampwidth == 2:
         typecode = "h"
     elif sampwidth == 4: