Update main.py

main.py

@@ -68,6 +68,7 @@ def run_pipeline(audio_path, src_lang="auto", tgt_lang="te",
         if (opt_fillers or opt_stutters) and word_segs:
             yield {"status": "processing", "step": 3, "message": "Step 3/5 — Removing fillers & stutters..."}
             import soundfile as sf
+            # Read the denoised audio — soundfile can read both WAV and MP3
             audio_data, sr = sf.read(clean1)
             if audio_data.ndim == 2:
                 audio_data = audio_data.mean(axis=1)
@@ -79,7 +80,10 @@ def run_pipeline(audio_path, src_lang="auto", tgt_lang="te",
             if opt_stutters:
                 audio_data, n_s = denoiser._remove_stutters(audio_data, sr, word_segs)
                 stats["stutters_removed"] = n_s
-            sf.write(clean1, audio_data, sr, subtype="PCM_24")
+            # Write to a fresh .wav — PCM_24 is WAV-only, never write to .mp3 path
+            clean_wav = os.path.join(out_dir, "clean_step3.wav")
+            sf.write(clean_wav, audio_data, sr, format="WAV", subtype="PCM_24")
+            clean1 = clean_wav  # downstream steps (Cloudinary upload) use this
         else:
             stats["fillers_removed"]  = 0
             stats["stutters_removed"] = 0