Update: timit_asr - make the dataset streamable

[lhoestq]

The TIMIT ASR dataset had two issues that was preventing it from being streamable:

1. it was missing a call to open before pd.read_csv
2. it was using os.path.dirname which is not supported for streaming

I made the dataset streamable by using open to load the CSV, and by adding the support for os.path.dirname in dataset scripts to stream data

You can now do

from datasets import load_dataset

timit_asr = load_dataset("timit_asr", streaming=True)
print(next(iter(timit_asr["train"])))

prints:

{"file": "zip://data/TRAIN/DR4/MMDM0/SI681.WAV::https://data.deepai.org/timit.zip",
"phonetic_detail": {"start": [0, 1960, 2466, 3480, 4000, 5960, 7480, 7880, 9400, 9960, 10680, 13480, 15680, 15880, 16920, 18297, 18882, 19480, 21723, 22516, 24040, 25190, 27080, 28160, 28560, 30120, 31832, 33240, 34640, 35968, 37720],
"utterance": ["h#", "w", "ix", "dcl", "s", "ah", "tcl", "ch", "ix", "n", "ae", "kcl", "t", "ix", "v", "r", "ix", "f", "y", "ux", "zh", "el", "bcl", "b", "iy", "y", "ux", "s", "f", "el", "h#"],
"stop": [1960, 2466, 3480, 4000, 5960, 7480, 7880, 9400, 9960, 10680, 13480, 15680, 15880, 16920, 18297, 18882, 19480, 21723, 22516, 24040, 25190, 27080, 28160, 28560, 30120, 31832, 33240, 34640, 35968, 37720, 39920]},
"sentence_type": "SI", "id": "SI681",
"speaker_id": "MMDM0",
"dialect_region": "DR4",
"text": "Would such an act of refusal be useful?",
"word_detail": {
    "start": [1960, 4000, 9400, 10680, 15880, 18297, 27080, 30120],
    "utterance": ["would", "such", "an", "act", "of", "refusal", "be", "useful"],
    "stop": [4000, 9400, 10680, 15880, 18297, 27080, 30120, 37720]
}}

cc @patrickvonplaten @vrindaprabhu