glam-imperial / emotionalconversionstargan Goto Github PK

Hi,Could you please write some detailed instructions on

1.Required datasets
2.Pre-processing data for training
3.How to train

Thanks so much.

Thank you!

I have run this code and planed to covert some samples,but I found the converted audios were noisy. Because I didn't change the code,so does "<path/to/model_checkpoint.ckpt>" mean "./checkpoints/model_step2/30000",when perform conversion.

Session1\dialog\transcriptions
Session1\dialog\MOCAP_rotated
Session1\dialog\MOCAP_head
Session1\dialog\MOCAP_hand
Session1\dialog\EmoEvaluation
which is the directory of Ses01F_impro01.txt?

Traceback (most recent call last):
File "/disk2/zhh/EmotionalConversionStarGAN-master/run_preprocessing.py", line 214, in
run_preprocessing(args)
File "/disk2/zhh/EmotionalConversionStarGAN-master/run_preprocessing.py", line 199, in run_preprocessing
generate_world_features(audio_filenames, data_dir)
File "/disk2/zhh/EmotionalConversionStarGAN-master/run_preprocessing.py", line 87, in generate_world_features
wav, labels = get_wav_and_labels(f, data_dir)
File "/disk2/zhh/EmotionalConversionStarGAN-master/utils/data_preprocessing_utils.py", line 111, in get_wav_and_labels
labels = concatenate_labels(category, speaker, dimensions, dimensions_dis)
UnboundLocalError: local variable 'dimensions_dis' referenced before assignment

how to solve this problem?

Hello.

I ran the same code as the dataset, but I have a problem with the pronunciation part.

Unlike the example voice you uploaded, your pronunciation is crushed and you hear a voice that seems to have drowned.

Please let me know if you know why this problem occurred.

Thank you.

For anyone trying to reproduce the results, edit the config files so that num_emos is 4, not 3. There are four emotions: angry, happy, sad, and neutral. Since the neutral emotion is assigned the label 3, if num_emos is 3, you will end up excluding the neutral emotion during training.