The viewer is disabled because this dataset repo requires arbitrary Python code execution. Please consider removing the loading script and relying on automated data support (you can use convert_to_parquet from the datasets library). If this is not possible, please open a discussion for direct help.

Librispeech-PC (punctuation and capitalization restored)

Original Dataset can be found in https://www.openslr.org/145/

  • I made it for personal use, so the code might not be pefect.

How to Use

  • Almost the same with Librispeech dataset module since i refered to its source code.

  • three types of transcripts are given

    • text_normalized : the trascript from Librispeech ASR
    • text, text_raw : the trascripts from Librispeech-PC
from datasets import load_dataset
from pprint import pprint
from IPython.display import display, Audio
import aiohttp
import os

# set huggingface cache directory for the extracted raw files and huggingface-cli token
# if don't, you might download the raw tar.gz file in home cache dir even if you set `cache_dir` param
os.environ['HF_HOME'] = "/data_dir/to/download"  
!export HF_HOME="/data_dir/to/download"


# download dataset
# if already have librispeech_asr in the cache_dir it will use the same audio files.
libripc = load_dataset("yoom618/librispeech_pc",
                       "all",  # all, clean, other
                       cache_dir="/data_dir/to/download",
                       trust_remote_code=True,
                       # storage_options={
                       #     # add if you need to increase the timeout for openslr download
                       #     'client_kwargs': {'timeout': aiohttp.ClientTimeout(total=7200)}
                       # },
                       )


# check dataset info
print(libripc)

display(Audio(libripc['train.clean.100'][0]['audio']['array'], 
              rate=libripc['train.clean.100'][0]['audio']['sampling_rate'],
              autoplay=False))
pprint(libripc['train.clean.100'][0]['audio'])
  • Data Sample

    {'audio': {'array': array([ 7.01904297e-04,  7.32421875e-04,  7.32421875e-04, ...,
                               -2.74658203e-04, -1.83105469e-04, -3.05175781e-05]),
               'path': '/data_dir/to/download/downloads/extracted/.../374-180298-0000.flac',
               'sampling_rate': 16000},
     'chapter_id': 180298,
     'duration': 14.529999732971191,
     'file': '/data_dir/to/download/downloads/extracted/.../374-180298-0000.flac',
     'id': '374-180298-0000',
     'speaker_id': 374,
     'text': 'Chapter sixteen I might have told you of the beginning of this '
             'liaison in a few lines, but I wanted you to see every step by which '
             'we came, I to agree to whatever Marguerite wished,',
     'text_normalized': 'CHAPTER SIXTEEN I MIGHT HAVE TOLD YOU OF THE BEGINNING OF '
                        'THIS LIAISON IN A FEW LINES BUT I WANTED YOU TO SEE EVERY '
                        'STEP BY WHICH WE CAME I TO AGREE TO WHATEVER MARGUERITE '
                        'WISHED',
     'text_raw': 'Chapter sixteen I might have told you of the beginning of this '
                 'liaison in a few lines, but I wanted you to see every step by '
                 'which we came, I to agree to whatever Marguerite wished,'}
    

The number of samples in Librispeech-PC

  • train
    • train.clean.100 : 26,041 ( 2,498 out of 28,539 were dropped )
    • train.clean.360 : 95,404 ( 8,610 out of 104,014 were dropped )
    • train.other.500 : 134,679 ( 14,009 out of 148,688 were dropped )
  • dev (validation)
    • dev.clean : 2,530 ( 173 out of 2,703 were dropped )
    • dev.other : 2,728 ( 136 out of 2,864 were dropped )
  • test
    • test.clean : 2,417 ( 203 out of 2,620 were dropped )
    • test.other : 2,856 ( 83 out of 2,939 were dropped )

Citation Information

@article{meister2023librispeechpc,
         title={LibriSpeech-PC: Benchmark for Evaluation of Punctuation and Capitalization Capabilities of end-to-end ASR Models}, 
         author={A. Meister and M. Novikov and N. Karpov and E. Bakhturina and V. Lavrukhin and B. Ginsburg},
         journal={arXiv preprint arXiv:2310.02943},
         year={2023},
        }
Downloads last month
50