What should I feed in ASR inference? about lingvo

[iamxiaoyubei]

It is the three feed ways:

hyps, src_frames, en_frames, scores = pred.Run(['hypotheses', 'src_frames', 'encoder_frames', 'scores'], wav='/tmp/librispeech/arctic_a0002.wav')

hyps, src_frames, en_frames, scores = pred.Run(['hypotheses', 'src_frames', 'encoder_frames', 'scores'], wav=wav_file)

hyps, src_frames, en_frames, scores = pred.Run(['hypotheses', 'src_frames', 'encoder_frames', 'scores'], wav=wav_tensor)

[iamxiaoyubei]

It is the codes to read waveform:

from scipy.io import wavfile
import tensorflow as tf
sample_rate, wav_file = wavfile.read('/tmp/librispeech/arctic_a0002.wav')
print(wav_file)
wav_tensor = tf.convert_to_tensor(wav_file)