diff --git a/output.wav b/output.wav
index 0069376..91bad2e 100644
Binary files a/output.wav and b/output.wav differ
diff --git a/sbv2_core/src/bert.rs b/sbv2_core/src/bert.rs
index 021d2bb..50762bf 100644
--- a/sbv2_core/src/bert.rs
+++ b/sbv2_core/src/bert.rs
@@ -25,11 +25,7 @@ pub fn predict(
     let output = outputs.get("output").unwrap();
 
     let content = output.try_extract_tensor::<f32>()?.to_owned();
-    println!("{:?}", content);
+    let (data, _) = content.clone().into_raw_vec_and_offset();
 
-    Ok(Array2::from_shape_vec(
-        (content.shape()[0], content.shape()[1]),
-        content.into_raw_vec(),
-    )
-    .unwrap())
+    Ok(Array2::from_shape_vec((content.shape()[0], content.shape()[1]), data).unwrap())
 }
diff --git a/sbv2_core/src/main.rs b/sbv2_core/src/main.rs
index 291e786..b2d9a0f 100644
--- a/sbv2_core/src/main.rs
+++ b/sbv2_core/src/main.rs
@@ -5,12 +5,10 @@ fn main() -> error::Result<()> {
     let text = "隣の客はよくかき食う客だ";
 
     let normalized_text = norm::normalize_text(text);
-    println!("{}", normalized_text);
 
     let jtalk = jtalk::JTalk::new()?;
     let (phones, tones, mut word2ph) = jtalk.g2p(&normalized_text)?;
     let (phones, tones, lang_ids) = nlp::cleaned_text_to_sequence(phones, tones);
-    println!("{:?}", phones);
 
     // add black
     let phones = utils::intersperse(&phones, 0);