An ASR model for transcribing laughter in Speech Laugh audio
path=/deepstore/datasets/hmi/speechlaugh-corpus # global data path- Using gdown to download the
.zipfile data and unzip it.
gdown 1VlQlyY3v3wtT2S047lwlTirWisz5mQ18 -O /path/to/data/switchboard.zip
#path=/deepstore/datasets/hmi/speechlaugh-corpus/switchboard_data # global datasets path
cd path/to/data #/deepstore/datasets/hmi/speechlaugh-corpus/switchboard_data
unzip switchboard.zip
# after unzip, the data will contain the following folders:
# - audio_wav
# - transcripts
- Generate audio_segments folder, this could be stored in the following path
path=/deepstore/datasets/hmi/speechlaugh-corpus/switchboard_data/audio_segments- Download the dataset from VocalSound and save it to
path/to/data/vocalsound_datafolder
wget -O vocalsound_16k.zip https://www.dropbox.com/s/c5ace70qh1vbyzb/vs_release_16k.zip?dl=1
#path=/deepstore/datasets/hmi/speechlaugh-corpus/vocalsound_data
unzip vocalsound_16k.zip- The path to the data would be:
path=/deepstore/datasets/hmi/speechlaugh-corpus/vocalsound_data/audio_16k- Download these datasets from HuggingFace datasets and saving to
data/huggingface_datafolder
- First set the path to HuggingFace cache to this folder
$ export HF_DATASETS_CACHE="../data/huggingface_data"
# or change to the global datasets
$ export HF_DATASETS_CACHE="/deepstore/datasets/hmi/speechlaugh-corpus/huggingface_data"
- Then download the datasets, given the dataset name in HuggingFace as follow:
- ami: "edinburghcstr/ami" "ihm" split="train"
- fsd50k_noisy: "sps44/fsdnoisy18k"
- audioset: "benjamin-paine/audio-set-16khz"