Convert CommonVoice tartar training file from mp3 to wav (8kHz), and remove unused file

[engineerchuan]

Used 'sox -i ' with default 14.4.2 options

soxi output for the wave file I created:

Input File     : 'assets/CommonVoice/cv-corpus-4-2019-12-10/tt/clips/common_voice_tt_00000000.wav'
Channels       : 1
Sample Rate    : 44100
Precision      : 16-bit
Duration       : 00:00:05.00 = 220500 samples = 375 CDDA sectors
File Size      : 441k
Bit Rate       : 706k
Sample Encoding: 16-bit Signed Integer PCM

I wasn't sure about whether we need to change Line 111 of torchaudio/datasets/commonvoice.py.

class COMMONVOICE(Dataset):
    """
    Create a Dataset for CommonVoice. Each item is a tuple of the form:
    (waveform, sample_rate, dictionary)
    where dictionary is a dictionary built from the tsv file with the following keys:
    client_id, path, sentence, up_votes, down_votes, age, gender, accent.
    """

    _ext_txt = ".txt"
    _ext_audio = ".mp3"
    _folder_audio = "clips"

[codecov]

Codecov Report

Merging #772 into master will not change coverage.
The diff coverage is n/a.

@@           Coverage Diff           @@
##           master     #772   +/-   ##
=======================================
  Coverage   89.53%   89.53%           
=======================================
  Files          32       32           
  Lines        2617     2617           
=======================================
  Hits         2343     2343           
  Misses        274      274           

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update c375490...9d0dc78. Read the comment docs.

[mthrok]

Hi @engineerchuan

Thanks for working on this.

Looks like waves_yesno/0_1_0_1_0_1_1_0.wav and one more file (not sure which one but the one used for test_gtzan) are actually used by test, which I overlooked. Could you put them ?

I wasn't sure about whether we need to change Line 111 of torchaudio/datasets/commonvoice.py.

We should be only changing test, so torchaudio/datasets/commonvoice.py should stay the same. Did you have an issue that way?

[mthrok]

Also, can you try reduce the sampling rate of the new wav file (assets/CommonVoice/cv-corpus-4-2019-12-10/tt/clips/common_voice_tt_00000000.wav.) to 8000 Hz? The quality of the audio should not matter to tests and we would like to keep them as small as possible.

[mthrok]

Looks good.

Thanks!

[mthrok]

Followup: Remove https://github.com/pytorch/audio/blob/master/test/test_datasets.py#L51 and https://github.com/pytorch/audio/blob/master/test/test_datasets.py#L49 to see if this test works on Windows.

@engineerchuan If you have time, would you like to try ^ ?

[engineerchuan]

Followup: Remove https://github.com/pytorch/audio/blob/master/test/test_datasets.py#L51 and https://github.com/pytorch/audio/blob/master/test/test_datasets.py#L49 to see if this test works on Windows.

@engineerchuan If you have time, would you like to try ^ ?

Yes I will try.