Data points remain tuples

[vincentqb]

Since this is what has been most used in domains, we keep tuples as the item to return for now.

for CommonVoice:

    # dictionary with keys:
    # client_id, path, sentence, up_votes, down_votes, age, gender, accent
    return waveform, sample_rate, dictionary

for LibriSpeech:

    return (
        waveform,
        sample_rate,
        utterance,
        int(speaker_id),
        int(chapter_id),
        int(utterance_id),
    )

for VCTK:

    return waveform, sample_rate, utterance, speaker_id, utterance_id

for YesNo:

    # labels = [int(c), ...]
    return waveform, sample_rate, labels

CC #303

[vincentqb]

Adding this as part of this PR to make sure that the file is closed.

[vincentqb]

Also converting to int though they come from file names.

[vincentqb]

Also converting to int though they come from file names.

[vincentqb]

Just modified CommonVoice to return a tuple (waveform, sample_rate, dictionary), where the dictionary contains the data and header from the tsv. This is closer to how the data is provided by CommonVoice. The only assumption on the tsv is that the second column is the name of the audio file to retrieve.

The previous iteration was:

    # waveform, sample_rate, client_id, path, sentence, up_votes, down_votes, age, gender, accent
    return(waveform, sample_rate, *line)

P.S. The notation with *line is not compatible with python 2.