File size: 5,306 Bytes
c8bff8b |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 |
import argparse
import pyarabic.araby as araby
from transformers import pipeline
from transformers.models.whisper.english_normalizer import BasicTextNormalizer
from datasets import load_dataset, Audio
import evaluate
from tqdm.auto import tqdm
wer_metric = evaluate.load("wer")
def is_target_text_in_range(ref):
if ref.strip() == "ignore time segment in scoring":
return False
else:
return ref.strip() != ""
def get_text(sample):
if "text" in sample:
return sample["text"]
elif "sentence" in sample:
return sample["sentence"]
elif "normalized_text" in sample:
return sample["normalized_text"]
elif "transcript" in sample:
return sample["transcript"]
elif "transcription" in sample:
return sample["transcription"]
else:
raise ValueError(
f"Expected transcript column of either 'text', 'sentence', 'normalized_text' or 'transcript'. Got sample of "
".join{sample.keys()}. Ensure a text column name is present in the dataset."
)
whisper_norm = BasicTextNormalizer()
def normalise(batch):
batch["norm_text"] = get_text(batch)
return batch
def remove_diacritics(batch):
batch["norm_text"] = araby.strip_diacritics(get_text(batch))
return batch
def data(dataset):
for i, item in enumerate(dataset):
yield {**item["audio"], "reference": item["norm_text"]}
def main(args):
batch_size = args.batch_size
whisper_asr = pipeline(
"automatic-speech-recognition", model=args.model_id, device=args.device
)
whisper_asr.model.config.forced_decoder_ids = (
whisper_asr.tokenizer.get_decoder_prompt_ids(
language=args.language, task="transcribe"
)
)
dataset = load_dataset(
args.dataset,
args.config,
split=args.split,
streaming=args.streaming,
use_auth_token=True,
)
# Only uncomment for debugging
if args.streaming:
dataset = dataset.take(args.max_eval_samples)
else:
if args.max_eval_samples is not None:
dataset = dataset.select(range(args.max_eval_samples))
dataset = dataset.cast_column("audio", Audio(sampling_rate=16000))
dataset = dataset.map(normalise)
dataset = dataset.filter(is_target_text_in_range, input_columns=["norm_text"])
predictions = []
references = []
# run streamed inference
if not args.streaming:
pbar = tqdm(total=len(dataset))
for out in whisper_asr(data(dataset), batch_size=batch_size):
pred = out["text"]
true = out["reference"][0]
if args.remove_diacritics:
pred = araby.strip_diacritics(pred)
true = araby.strip_diacritics(true)
if args.normalise:
pred = whisper_norm(pred)
true = whisper_norm(true)
predictions.append(pred)
references.append(true)
if not args.streaming:
pbar.update(1)
if not args.streaming:
pbar.close()
wer = wer_metric.compute(references=references, predictions=predictions)
wer = round(100 * wer, 2)
print("WER:", wer)
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument(
"--model_id",
type=str,
required=True,
help="Model identifier. Should be loadable with 🤗 Transformers",
)
parser.add_argument(
"--dataset",
type=str,
default="mozilla-foundation/common_voice_11_0",
help="Dataset name to evaluate the `model_id`. Should be loadable with 🤗 Datasets",
)
parser.add_argument(
"--config",
type=str,
required=True,
help="Config of the dataset. *E.g.* `'en'` for the English split of Common Voice",
)
parser.add_argument(
"--split",
type=str,
default="test",
help="Split of the dataset. *E.g.* `'test'`",
)
parser.add_argument(
"--device",
type=int,
default=-1,
help="The device to run the pipeline on. -1 for CPU (default), 0 for the first GPU and so on.",
)
parser.add_argument(
"--batch_size",
type=int,
default=16,
help="Number of samples to go through each streamed batch.",
)
parser.add_argument(
"--max_eval_samples",
type=int,
default=None,
help="Number of samples to be evaluated. Put a lower number e.g. 64 for testing this script.",
)
parser.add_argument(
"--streaming",
default=False,
action="store_true",
help="Choose whether you'd like to download the entire dataset or stream it during the evaluation.",
)
parser.add_argument(
"--language",
type=str,
required=True,
help="Two letter language code for the transcription language, e.g. use 'en' for English.",
)
parser.add_argument(
"--remove_diacritics",
default=False,
action="store_true",
help="Choose whether you'd like remove_diacritics",
)
parser.add_argument(
"--normalise",
default=False,
action="store_true",
help="Choose whether you'd like whisper norm",
)
args = parser.parse_args()
main(args)
|