diff --git a/paddlespeech/t2s/exps/fastspeech2/preprocess.py b/paddlespeech/t2s/exps/fastspeech2/preprocess.py
index 5bda75451b071321e681adceb598f29162b5fb8c..db1842b2e89fe3044e96ca4babb07c1796d06da3 100644
--- a/paddlespeech/t2s/exps/fastspeech2/preprocess.py
+++ b/paddlespeech/t2s/exps/fastspeech2/preprocess.py
@@ -86,6 +86,9 @@ def process_sentence(config: Dict[str, Any],
         logmel = mel_extractor.get_log_mel_fbank(wav)
         # change duration according to mel_length
         compare_duration_and_mel_length(sentences, utt_id, logmel)
+        # utt_id may be popped in compare_duration_and_mel_length
+        if utt_id not in sentences:
+            return None
         phones = sentences[utt_id][0]
         durations = sentences[utt_id][1]
         num_frames = logmel.shape[0]
diff --git a/paddlespeech/t2s/exps/speedyspeech/preprocess.py b/paddlespeech/t2s/exps/speedyspeech/preprocess.py
index 3f81c4e14753d19e51db5d23f5e75440c67de34b..e833d13940530f293842a842b65f33cf6d03d9bd 100644
--- a/paddlespeech/t2s/exps/speedyspeech/preprocess.py
+++ b/paddlespeech/t2s/exps/speedyspeech/preprocess.py
@@ -79,6 +79,9 @@ def process_sentence(config: Dict[str, Any],
         logmel = mel_extractor.get_log_mel_fbank(wav)
         # change duration according to mel_length
         compare_duration_and_mel_length(sentences, utt_id, logmel)
+        # utt_id may be popped in compare_duration_and_mel_length
+        if utt_id not in sentences:
+            return None
         labels = sentences[utt_id][0]
         # extract phone and duration
         phones = []
diff --git a/paddlespeech/t2s/exps/tacotron2/preprocess.py b/paddlespeech/t2s/exps/tacotron2/preprocess.py
index 7f41089ebf9d71b336d082b065e8b50c541f7edd..14a0d7eae227f5650a716bc656f3d0c32ee077e3 100644
--- a/paddlespeech/t2s/exps/tacotron2/preprocess.py
+++ b/paddlespeech/t2s/exps/tacotron2/preprocess.py
@@ -82,6 +82,9 @@ def process_sentence(config: Dict[str, Any],
         logmel = mel_extractor.get_log_mel_fbank(wav)
         # change duration according to mel_length
         compare_duration_and_mel_length(sentences, utt_id, logmel)
+        # utt_id may be popped in compare_duration_and_mel_length
+        if utt_id not in sentences:
+            return None
         phones = sentences[utt_id][0]
         durations = sentences[utt_id][1]
         num_frames = logmel.shape[0]