Update app.py

app.py

@@ -91,45 +91,24 @@ def encode_decode_focal(audio_input):
         with torch.no_grad():
             toks = codec.sig_to_toks(sig)
             rec_sig = codec.toks_to_sig(toks)
+            
+            # Get binary codes for true compression
             codes = codec.toks_to_codes(toks)
         
-        # --- DEBUG: Print what we actually have ---
-        print(f"\n=== DEBUG INFO ===")
-        print(f"Tokens shape: {toks.shape}")
-        print(f"Tokens dtype: {toks.dtype}")
-        print(f"Tokens min/max: {toks.min().item()}/{toks.max().item()}")
-        print(f"First 10 tokens: {toks[0, :10].cpu().numpy()}")
-        
-        print(f"\nCodes shape: {codes.shape}")
-        print(f"Codes dtype: {codes.dtype}")
-        print(f"Codes min/max: {codes.min().item()}/{codes.max().item()}")
-        print(f"First 10 codes (flattened): {codes.flatten()[:10].cpu().numpy()}")
-        print(f"=================\n")
-        
-        # --- Save the compressed data ---
+       # --- Save the compressed tokens to a temporary .fc file ---
         temp_dir = tempfile.mkdtemp()
         fc_file_path = os.path.join(temp_dir, "compressed_tokens.fc")
-        
-        # Try saving tokens directly as integers (more reliable)
-        toks_cpu = toks.cpu().numpy().astype(np.int32)
-        
+
+        # Save as raw binary data (just the token values)
+        toks_cpu = toks.cpu().numpy().astype(np.int16)  # Convert to numpy
         with open(fc_file_path, 'wb') as f:
-            # Write header: number of tokens
-            num_tokens = toks_cpu.size
-            f.write(num_tokens.to_bytes(4, byteorder='little'))
-            # Write tokens as raw bytes
-            f.write(toks_cpu.tobytes())
-        
-        # Calculate stats
+            f.write(toks_cpu.tobytes())  # Write raw bytes
+
         file_size_bytes = os.path.getsize(fc_file_path)
         duration_sec = sig.shape[-1] / codec.sample_rate_input
-        expected_size = (160 * duration_sec) / 8
-        actual_bitrate = (file_size_bytes * 8) / duration_sec
-        
-        print(f"Duration: {duration_sec:.2f}s")
-        print(f"Num tokens: {num_tokens}")
+        expected_size = (160 * duration_sec) / 8  # 160 bits/sec → bytes
+        print(f"Tokens saved to {fc_file_path}")
         print(f"File size: {file_size_bytes} bytes (expected: ~{expected_size:.0f} bytes)")
-        print(f"Actual bitrate: {actual_bitrate:.0f} bps")
         
         # Move audio back to CPU
         decoded_wav_output = rec_sig.cpu().numpy().squeeze()
@@ -137,7 +116,7 @@ def encode_decode_focal(audio_input):
         if len(decoded_wav_output.shape) == 0:
             decoded_wav_output = decoded_wav_output.reshape(1)
         
-        status_msg = f"✅ Duration: {duration_sec:.1f}s | Tokens: {num_tokens} | File: {file_size_bytes} bytes | Bitrate: {actual_bitrate:.0f} bps"
+        status_msg = f"✅ Duration: {duration_sec:.1f}s | File: {file_size_bytes} bytes | Bitrate: {actual_bitrate:.0f} bps"
         
         return (codec.sample_rate_output, decoded_wav_output), fc_file_path, status_msg