ayameRushia
/

wav2vec2-large-xls-r-300m-ia

Automatic Speech Recognition Transformers PyTorch

Interlingua wav2vec2 generated_from_trainer hf-asr-leaderboard robust-speech-event mozilla-foundation/common_voice_8_0 Eval Results Inference Endpoints

Model card Files Files and versions Community

wav2vec2-large-xls-r-300m-ia / eval.py

ayameRushia

Upload eval.py

af1d29d about 2 years ago

raw history blame contribute delete

No virus

6.28 kB

	from datasets import load_dataset, load_metric, Audio, Dataset
	from transformers import pipeline, AutoFeatureExtractor, AutoTokenizer, AutoConfig, AutoModelForCTC, Wav2Vec2Processor, Wav2Vec2ProcessorWithLM
	import re
	import torch
	import argparse
	from typing import Dict

	def log_results(result: Dataset, args: Dict[str, str]):
	""" DO NOT CHANGE. This function computes and logs the result metrics. """

	log_outputs = args.log_outputs
	dataset_id = "_".join(args.dataset.split("/") + [args.config, args.split])

	# load metric
	wer = load_metric("wer")
	cer = load_metric("cer")

	# compute metrics
	wer_result = wer.compute(references=result["target"], predictions=result["prediction"])
	cer_result = cer.compute(references=result["target"], predictions=result["prediction"])

	# print & log results
	result_str = (
	f"WER: {wer_result}\n"
	f"CER: {cer_result}"
	)
	print(result_str)

	with open(f"{dataset_id}_eval_results.txt", "w") as f:
	f.write(result_str)

	# log all results in text file. Possibly interesting for analysis
	if log_outputs is not None:
	pred_file = f"log_{dataset_id}_predictions.txt"
	target_file = f"log_{dataset_id}_targets.txt"

	with open(pred_file, "w") as p, open(target_file, "w") as t:

	# mapping function to write output
	def write_to_file(batch, i):
	p.write(f"{i}" + "\n")
	p.write(batch["prediction"] + "\n")
	t.write(f"{i}" + "\n")
	t.write(batch["target"] + "\n")

	result.map(write_to_file, with_indices=True)


	def normalize_text(text: str) -> str:
	"""DO ADAPT FOR YOUR USE CASE. this function normalizes the target text."""

	CHARS_TO_IGNORE = [",", "?", "¿", ".", "!", "¡", ";", "；", ":", '""', "%", '"', "�", "ʿ", "·", "჻", "~", "՞",
	"؟", "،", "।", "॥", "«", "»", "„", "“", "”", "「", "」", "‘", "’", "《", "》", "(", ")", "[", "]",
	"{", "}", "=", "`", "_", "+", "<", ">", "…", "–", "°", "´", "ʾ", "‹", "›", "©", "®", "—", "→", "。",
	"、", "﹂", "﹁", "‧", "～", "﹏", "，", "｛", "｝", "（", "）", "［", "］", "【", "】", "‥", "〽",
	"『", "』", "〝", "〟", "⟨", "⟩", "〜", "：", "！", "？", "♪", "؛", "/", "\\", "º", "−", "^", "ʻ", "ˆ"]
	chars_to_ignore_regex = f"[{re.escape(''.join(CHARS_TO_IGNORE))}]"

	text = re.sub(chars_to_ignore_regex, "", text.lower())

	# In addition, we can normalize the target text, e.g. removing new lines characters etc...
	# note that order is important here!
	token_sequences_to_ignore = ["\n\n", "\n", " ", " "]

	for t in token_sequences_to_ignore:
	text = " ".join(text.split(t))

	return text


	def main(args):
	# load dataset
	dataset = load_dataset(args.dataset, args.config, split=args.split, use_auth_token=True)

	# for testing: only process the first two examples as a test
	# dataset = dataset.select(range(10))

	# load processor
	if args.greedy:
	processor = Wav2Vec2Processor.from_pretrained(args.model_id)
	decoder = None
	else:
	processor = Wav2Vec2ProcessorWithLM.from_pretrained(args.model_id)
	decoder = processor.decoder

	feature_extractor = processor.feature_extractor
	tokenizer = processor.tokenizer

	# resample audio
	dataset = dataset.cast_column("audio", Audio(sampling_rate=feature_extractor.sampling_rate))

	# load eval pipeline
	if args.device is None:
	args.device = 0 if torch.cuda.is_available() else -1

	config = AutoConfig.from_pretrained(args.model_id)
	model = AutoModelForCTC.from_pretrained(args.model_id)

	#asr = pipeline("automatic-speech-recognition", model=args.model_id, device=args.device)
	asr = pipeline("automatic-speech-recognition", config=config, model=model, tokenizer=tokenizer,
	feature_extractor=feature_extractor, decoder=decoder, device=args.device)


	# map function to decode audio
	def map_to_pred(batch, args=args, asr=asr):
	prediction = asr(batch["audio"]["array"], chunk_length_s=args.chunk_length_s, stride_length_s=args.stride_length_s)

	batch["prediction"] = prediction["text"]
	batch["target"] = normalize_text(batch["sentence"])
	return batch

	# run inference on all examples
	result = dataset.map(map_to_pred, remove_columns=dataset.column_names)

	# filtering out empty targets
	result = result.filter(lambda example: example["target"] != "")

	# compute and log_results
	# do not change function below
	log_results(result, args)


	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, required=True, 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 Common Voice"
	)
	parser.add_argument(
	"--split", type=str, required=True, help="Split of the dataset. E.g. `'test'`"
	)
	parser.add_argument(
	"--chunk_length_s", type=float, default=None, help="Chunk length in seconds. Defaults to None. For long audio files a good value would be 5.0 seconds."
	)
	parser.add_argument(
	"--stride_length_s", type=float, default=None, help="Stride of the audio chunks. Defaults to None. For long audio files a good value would be 1.0 seconds."
	)
	parser.add_argument(
	"--log_outputs", action='store_true', help="If defined, write outputs to log file for analysis."
	)
	parser.add_argument(
	"--greedy", action='store_true', help="If defined, the LM will be ignored during inference."
	)
	parser.add_argument(
	"--device",
	type=int,
	default=None,
	help="The device to run the pipeline on. -1 for CPU (default), 0 for the first GPU and so on.",
	)
	args = parser.parse_args()

	main(args)