Update gradio_app.py

gradio_app.py

@@ -53,10 +53,11 @@ def separate_speakers_core(audio_path):
     output_files = []
     for i in range(ests_speech.shape[0]):
         path = os.path.join(output_dir, f"speaker_{i+1}.wav")
-        waveform = ests_speech[i].unsqueeze(0).cpu()  # shape: (1, samples)
-        torchaudio.save(path, waveform, TARGET_SR)    # Saves as valid WAV
+        waveform = ests_speech[i].cpu().unsqueeze(0)  # (1, samples)
+        torchaudio.save(path, waveform, TARGET_SR)
         output_files.append(path)
 
+
     return output_files
 
 
@@ -133,12 +134,8 @@ def separate_speakers_video(video_path):
 
     output_videos = []
     for i, audio_file in enumerate(output_files):
-        # Re-encode to ensure ffmpeg/moviepy compatibility
-        reencoded_path = os.path.join(output_dir, f"speaker_{i+1}_final.wav")
-        convert_to_ffmpeg_friendly(audio_file, reencoded_path)
-
         speaker_video_path = os.path.join(output_dir, f"speaker_{i+1}_video.mp4")
-        video_with_sep_audio = attach_audio_to_video(video, reencoded_path, speaker_video_path)
+        video_with_sep_audio = attach_audio_to_video(video, audio_file, speaker_video_path)
         output_videos.append(video_with_sep_audio)
 
     updates = []
@@ -152,6 +149,7 @@ def separate_speakers_video(video_path):
 
 
 
+
 # --- Gradio UI ---
 with gr.Blocks() as demo:
     gr.Markdown("# TIGER: Time-frequency Interleaved Gain Extraction and Reconstruction for Efficient Speech Separation")