Fix: Improve spectrogram rendering stability and accuracy

Replaced manual time scaling with librosa.time_to_frames to prevent the visualization from freezing at the end of the video. Added a safety check for empty spectrograms to avoid crashes with very short audio clips.

app.py

@@ -505,8 +505,6 @@ def process_audio_to_video(*args, progress=gr.Progress(track_tqdm=True)):
 
             text_clips.append(txt_clip)
 
-
-        
         N_FFT, HOP_LENGTH, N_BANDS = 2048, 512, 32
         MIN_DB, MAX_DB = -80.0, 0.0
 
@@ -529,8 +527,22 @@ def process_audio_to_video(*args, progress=gr.Progress(track_tqdm=True)):
             if not image_clips:
                 for i in range(1, 9):
                     y_pos = int(i * (HEIGHT / 9)); frame[y_pos-1:y_pos, :] = grid_rgb
-
-            time_idx = min(int((t / duration) * S_mel_db.shape[1]), S_mel_db.shape[1] - 1)
+            
+            # 1. Safety Check: If the spectrogram has no time frames (e.g., from an extremely short audio file),
+            #    return a blank frame immediately to prevent an IndexError.
+            if S_mel_db.shape[1] == 0:
+                return frame
+
+            # 2. Use librosa.time_to_frames to accurately convert the video time `t`
+            #    into a spectrogram frame index. This is far more reliable than manual scaling
+            #    and solves the problem of missing content on the rightmost side of the video.
+            time_idx = librosa.time_to_frames(t, sr=current_sr, hop_length=HOP_LENGTH)
+            
+            # 3. Boundary Protection: Although time_to_frames is accurate, this extra `min`
+            #    call acts as a safeguard to ensure the index never exceeds the array's
+            #    maximum valid index, preventing any edge-case errors.
+            time_idx = min(time_idx, S_mel_db.shape[1] - 1)
+            
             bar_width = WIDTH / N_BANDS
             for i in range(N_BANDS):
                 energy_db = S_mel_db[i, time_idx]