mtkresearch / clairaudience Goto Github PK

Hi, I wnat to train model use GigaSpeech now, the Guidance doc said ensure that Gigaspeech medium is downladed to data/hf_dd_data/gigaspeech/m，i don’t know which format is available，origin long wav ? or short wav slice according to segments (timestamp json file)?
moreover，what means of dataset_name in func setup_dataset (data_process.py);
line 94: elif dataset_name == gigaspeech:
line 101: elif dataset_name == gigaspeech_extracted:
line 115: elif dataset_name == gigaspeech:

thx！

I see the cfg_gigaspeech_ft_base.json in the [config] folder. I think that is the setting for audio-text training.
Can you provide the setting for text-only training? I would appreciate it very much.