diff --git a/tools/extract_embedding.py b/tools/extract_embedding.py
index cb198cb..de48779 100755
--- a/tools/extract_embedding.py
+++ b/tools/extract_embedding.py
@@ -26,9 +26,9 @@ def single_job(utt):
     if sample_rate != 16000:
         audio = torchaudio.transforms.Resample(orig_freq=sample_rate, new_freq=16000)(audio)
     feat = kaldi.fbank(audio,
-                        num_mel_bins=80,
-                        dither=0,
-                        sample_frequency=16000)
+                       num_mel_bins=80,
+                       dither=0,
+                       sample_frequency=16000)
     feat = feat - feat.mean(dim=0, keepdim=True)
     embedding = ort_session.run(None, {ort_session.get_inputs()[0].name: feat.unsqueeze(dim=0).cpu().numpy()})[0].flatten().tolist()
     return utt, embedding
diff --git a/tools/extract_speech_token.py b/tools/extract_speech_token.py
index 2829624..26aa296 100755
--- a/tools/extract_speech_token.py
+++ b/tools/extract_speech_token.py
@@ -33,7 +33,7 @@ def single_job(utt):
     else:
         feat = whisper.log_mel_spectrogram(audio, n_mels=128)
         speech_token = ort_session.run(None, {ort_session.get_inputs()[0].name: feat.detach().cpu().numpy(),
-                                                ort_session.get_inputs()[1].name: np.array([feat.shape[2]], dtype=np.int32)})[0].flatten().tolist()
+                                              ort_session.get_inputs()[1].name: np.array([feat.shape[2]], dtype=np.int32)})[0].flatten().tolist()
     return utt, speech_token