data issue

[wrz1999]

Hello, I would like to know where should I put the wjs0-mix dataset in order to generate the RIR dataset？

[quancs]

Hello, the mixture signals are dynamically generated with rirs and speech signals. So, to generate the rirs, you don't need to put wsj0-mix somewhere. Acctually, to generate the dataset used in our papers, you need the wsj0 dataset (specify its path in datamodule, i.e. the 'data' in config file) and the generated rirs (also config it in the data item of config file).

[wrz1999]

Is the path of the final dataset wsj0-mix modified in files such as NB-BLSTM-fit.yaml? clean_speech_dataset: wsj0-mix clean_speech_dir: /dev/shm/quancs/ rir_dir: /dev/shm/quancs/rir_cfg_4 Is clean_speech_dir the wsj0-mix path?

[quancs]

clean speech path = clean_speech_dir + the path in NBSS/configs/wsj0-mix/mix_2_spk_tr.txt or other txt.

clean_speech_dataset is just a name for code to understand how to generate the prepare the dataset.

Is the path of the final dataset wsj0-mix modified in files such as NB-BLSTM-fit.yaml?

the mixture signals are dynamically genderated and we don't store them. thus there is no such path for final dataset in config file.

Is clean_speech_dir the wsj0-mix path?

it's not. its function is as mentioned before

[wrz1999]

Can I directly train the model by changing the clean speech path and generating RIR data?

[quancs]

Can I directly train the model by changing the clean speech path and generating RIR data?

yes. generate the rirs first. then change the pathes in config file. after that, you can directly train the model.

[wrz1999]

Thank you very much for your answer. Happy New Year to you in advance.

[quancs]

You are welcome. Happy New Year too. ^_^