feat: add wav_to_mp3 conversion and update return values to provide MP3 output files

app.py

@@ -127,6 +127,23 @@ def ensure_wav(filepath: str) -> str:
     return filepath
 
 
+def wav_to_mp3(wav_path: str) -> str:
+    """Convert temporary WAV output to MP3 using ffmpeg for smaller file size."""
+    mp3_path = wav_path.rsplit(".", 1)[0] + ".mp3"
+    subprocess.run(
+        ["ffmpeg", "-y", "-i", wav_path, "-vn", "-ar", "44100", "-ac", "1", "-b:a", "128k", mp3_path],
+        check=True,
+        stdout=subprocess.DEVNULL,
+        stderr=subprocess.DEVNULL
+    )
+    # Remove the temporary wav file
+    try:
+        os.remove(wav_path)
+    except Exception as e:
+        logger.warning(f"Failed to remove temp WAV file {wav_path}: {e}")
+    return mp3_path
+
+
 import torchaudio.functional as F_audio
 
 def post_process_audio(waveform: Tensor, sr: int) -> Tensor:
@@ -222,19 +239,23 @@ def demo_fn(
     enhanced_wav = tempfile.NamedTemporaryFile(suffix="enhanced.wav", delete=False).name
     save_audio(enhanced_wav, enhanced, sr)
 
-    logger.info(f"saved audios: {noisy_wav}, {enhanced_wav}")
+    logger.info("Converting outputs to MP3...")
+    noisy_mp3 = wav_to_mp3(noisy_wav)
+    enhanced_mp3 = wav_to_mp3(enhanced_wav)
+
+    logger.info(f"saved audios: {noisy_mp3}, {enhanced_mp3}")
 
     ax_noisy.clear()
     ax_enh.clear()
     noisy_im = spec_im(sample, sr=sr, figure=fig_noisy, ax=ax_noisy)
     enh_im = spec_im(enhanced, sr=sr, figure=fig_enh, ax=ax_enh)
 
-    filter = [speech_upl, noisy_wav, enhanced_wav]
+    filter = [speech_upl, noisy_mp3, enhanced_mp3]
     if mic_input is not None and mic_input != "":
         filter.append(mic_input)
     cleanup_tmp(filter)
 
-    return noisy_wav, noisy_im, enhanced_wav, enh_im
+    return noisy_mp3, noisy_im, enhanced_mp3, enh_im
 
 def specshow(
     spec,