search

HPI-DeepLearning / crnn-lid

Code for the paper Language Identification Using Deep Convolutional Recurrent Neural Networks
GNU General Public License v3.0
105 stars 48 forks source link

Changing to shorter segments #38

Open Themba4Sho opened 2 years ago

Themba4Sho commented 2 years ago

What do I need to change to train on shorter segments? I used data with minimum length of 3 seconds, but the wav_to_spec module still processes speech with only 10 seconds or more.

Vadim2S commented 1 year ago

In case you still want answer: 1) Go to config.yaml. 2) See input shape parameter like [129, 500, 1]. 3) Set segment_length as 3. This is minimal audio lenght in seconds 4) Set pixel_per_second = 500 / segment_length. In general you must provide input_shape[1] = segment_length * pixel_per_second by any means you wish.