How does the type of audio affect the predictive charsiu aligner. The audio is in wav format

[DIKSHAAGARWAL2015]

[lingjzhu]

The training data is mostly clean speech, so non-speech audio effects probably will make it even more error-prone. You can test it yourself.

[DIKSHAAGARWAL2015]

When I use my audio's, I get sampling assertion and size error. Expected 2D (unbatched) or 3D (batched) input to conv1d, but got input of size: [1, 1, 1046752, 2]

[lingjzhu]

The sampling rate should be 16k and the audio should be mono channel. The input size seems to suggest that your audio is a multi-channel one.

[DIKSHAAGARWAL2015]

The audio becomes multi channel since the data is not lab prepared and directly recorded using micro phone. How I ensure mono channel by using transformations ?

[lingjzhu]

You can check out librosa: https://librosa.org/doc/main/generated/librosa.load.html

[DIKSHAAGARWAL2015]

When it predicts phonemes to print the corresponding words in audio do I need to add print statements?

[lingjzhu]

It only gives you phonemes, not words.