diff --git a/scripts/process_dataset.py b/scripts/process_dataset.py
index 9b43f30..7f1d00a 100644
--- a/scripts/process_dataset.py
+++ b/scripts/process_dataset.py
@@ -2,7 +2,7 @@ import os
 import json
 import torch
 import torchaudio
-
+import numpy as np
 from tqdm.auto import tqdm
 from pathlib import Path
 from vall_e.config import cfg
@@ -61,12 +61,12 @@ for dataset_name in sorted(os.listdir(f'./{input_audio}/')):
 
 				if cfg.inference.audio_backend == "dac":
 					np.save(open(_replace_file_extension(outpath, audio_extension), "wb"), {
-						"codes": qnt.codes.numpy().astype(np.uint16),
+						"codes": qnt.codes.cpu().numpy().astype(np.uint16),
 						"metadata": {
 							"original_length": qnt.original_length,
 							"sample_rate": qnt.sample_rate,
 							
-							"input_db": qnt.input_db.numpy().astype(np.float32),
+							"input_db": qnt.input_db.cpu().numpy().astype(np.float32),
 							"chunk_length": qnt.chunk_length,
 							"channels": qnt.channels,
 							"padding": qnt.padding,
@@ -75,7 +75,7 @@ for dataset_name in sorted(os.listdir(f'./{input_audio}/')):
 					})
 				else:
 					np.save(open(_replace_file_extension(outpath, audio_extension), "wb"), {
-						"codes": qnt.numpy().astype(np.uint16),
+						"codes": qnt.cpu().numpy().astype(np.uint16),
 						"metadata": {
 							"original_length": waveform.shape[-1],
 							"sample_rate": sample_rate,
@@ -186,12 +186,12 @@ for dataset_name in sorted(os.listdir(f'./{input_audio}/')):
 
 					if cfg.inference.audio_backend == "dac":
 						np.save(open(_replace_file_extension(outpath, audio_extension), "wb"), {
-							"codes": qnt.codes.numpy().astype(np.uint16),
+							"codes": qnt.codes.cpu().numpy().astype(np.uint16),
 							"metadata": {
 								"original_length": qnt.original_length,
 								"sample_rate": qnt.sample_rate,
 								
-								"input_db": qnt.input_db.numpy().astype(np.float32),
+								"input_db": qnt.input_db.cpu().numpy().astype(np.float32),
 								"chunk_length": qnt.chunk_length,
 								"channels": qnt.channels,
 								"padding": qnt.padding,
@@ -204,7 +204,7 @@ for dataset_name in sorted(os.listdir(f'./{input_audio}/')):
 						})
 					else:
 						np.save(open(_replace_file_extension(outpath, audio_extension), "wb"), {
-							"codes": qnt.numpy().astype(np.uint16),
+							"codes": qnt.cpu().numpy().astype(np.uint16),
 							"metadata": {
 								"original_length": waveform.shape[-1],
 								"sample_rate": sample_rate,