Update app.py

app.py

@@ -26,26 +26,77 @@ def save_as_word(text, filename="merged_transcripts.docx"):
 def convert_to_wav_if_needed(input_path):
     """
     If the input file is not WAV, try to convert it to WAV using pydub/ffmpeg.
+    First attempts pydub (ffmpeg via AudioSegment). If that fails (common for obscure
+    .dct variants), it falls back to trying several ffmpeg raw-format heuristics
+    (different sample rates, signed/unsigned 8/16-bit) until one succeeds.
     Returns path to WAV file (may be same as input if already WAV).
+    Raises an exception with ffmpeg stderr if no conversion worked.
     """
+    import subprocess
+
     lower = input_path.lower()
     if lower.endswith('.wav'):
         return input_path
 
-    # create a temp wav file
+    # 1) Try pydub automatic detection first (simplest)
     tmp_wav = tempfile.NamedTemporaryFile(suffix='.wav', delete=False)
     tmp_wav.close()
     try:
-        # pydub will use ffmpeg under the hood
         AudioSegment.from_file(input_path).export(tmp_wav.name, format='wav')
         return tmp_wav.name
-    except Exception as e:
-        # cleanup if conversion failed
+    except Exception as e_auto:
+        # Clean up the partial file
         try:
             os.unlink(tmp_wav.name)
         except Exception:
             pass
-        raise e
+        # We'll try a set of ffmpeg heuristics below
+        ffmpeg_errors = []
+
+    # 2) Fallback: try various raw-format guesses with ffmpeg
+    guesses = [
+        # fmt, sample_rate, channels
+        ('s16le', 16000, 1),
+        ('s16le', 8000, 1),
+        ('u8', 8000, 1),
+        ('s16le', 44100, 1),
+        ('s16le', 16000, 2),
+        ('adpcm_ima_wav', 8000, 1),
+    ]
+
+    for fmt, sr, ch in guesses:
+        tmp_wav = tempfile.NamedTemporaryFile(suffix='.wav', delete=False)
+        tmp_wav.close()
+        cmd = [
+            'ffmpeg', '-y', '-f', fmt, '-ar', str(sr), '-ac', str(ch), '-i', input_path,
+            tmp_wav.name
+        ]
+        try:
+            proc = subprocess.run(cmd, capture_output=True, text=True, timeout=60)
+        except Exception as e_run:
+            ffmpeg_errors.append(f"ffmpeg run failed for fmt={fmt},sr={sr},ch={ch}: {e_run}")
+            try:
+                os.unlink(tmp_wav.name)
+            except Exception:
+                pass
+            continue
+
+        if proc.returncode == 0 and os.path.exists(tmp_wav.name) and os.path.getsize(tmp_wav.name) > 100:
+            # success
+            return tmp_wav.name
+        else:
+            err = proc.stderr or proc.stdout or 'no ffmpeg output'
+            ffmpeg_errors.append(f"fmt={fmt},sr={sr},ch={ch} -> rc={proc.returncode} -> {err}")
+            try:
+                os.unlink(tmp_wav.name)
+            except Exception:
+                pass
+
+    # If we get here, nothing worked. Provide the accumulated errors.
+    msg = "Could not convert file to WAV. Tried pydub and several ffmpeg heuristics. Errors:
+" + "
+".join(ffmpeg_errors)
+    raise Exception(msg)
 
 
 def transcribe_multiple(file_paths, model_name, advanced, merge_checkbox, zip_file=None, zip_password=None):