Merge pull request #1660 from yt605155624/fix_pre

[TTS]fix preprocess bug, test=tts

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]

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 = []

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]