size = random.choice(range(min_size, max_size + 1))
tokens = tokenize(subtitle)
while len(tokens) < size:
tokens = tokenize(next(subtitles_it))
start = random.randint(0, len(tokens) - size)
yield " ".join(t.value.lower() for t in tokens[start:start + size])
After Change
def get_noise_iterator(language, min_size, max_size):
subtitles = get_subtitles(language)
tokenized_subtitles = [tokenize(s) for s in subtitles]
tokenized_subtitles = [t for t in tokenized_subtitles if
len(t) >= max_size]
if len(tokenized_subtitles) == 0:
raise ValueError("Could not find long enought subtitles")
subtitles_it = cycle(np.random.permutation(tokenized_subtitles))
for tokens in subtitles_it:
