update speech.dtypeg

gradio_asr_en_libri100_word_vs_bpe.py

@@ -54,6 +54,7 @@ def predict(wav_file):
     logger.info(f'speech.shape: {speech.shape}')
     logger.info(f'speech.dtype: {speech.dtype}')
     logger.info(f'speech rate: {rate}')
+    speech = speech.astype('float32')
     
     # number_of_samples = round(len(speech_raw) * float(16000) / rate)
     # speech = sps.resample(speech_raw, number_of_samples)