Dataset viewer won't play/stream m4a audio files

馃Datasets
1

Dataset viewer throws LibsndfileError for m4a files. Here鈥檚 my dataset for reference -
ProgramComputer/test 路 Datasets at Hugging Face. Upon looking at the traceback, I think soundfile can鈥檛 read m4a files.

Below, I tested librosa and sndfile; librosa was able to play m4a files.

import io
from urllib.request import urlopen
import librosa
import pydub
from IPython.display import Audio
import soundfile as sf
import pathlib


wav = io.BytesIO()

url = "https://github.com/robovm/apple-ios-samples/raw/master/avTouch/sample.m4a"
z = io.BytesIO(urlopen(url).read())
pathlib.Path('./sneeze.m4a').write_bytes(z.getbuffer())
y, sr = librosa.load('./sneeze.m4a')

Audio(data=y, rate=sr)

data, samplerate = sf.read('./sneeze.m4a')
Audio(data=data, rate=samplerate)
2

The m4a support in libsndfile (soundfile is a Python binding to it) has been requested here (the discussion seems to be active again). In the meantime, you can fix the viewer using librosa inside a dataset script to decode the files, and return decoded NumPy arrays as values.