eval results

eval.py

@@ -0,0 +1,128 @@
+#!/usr/bin/env python3
+import argparse
+import re
+from typing import Dict
+
+from datasets import Audio, Dataset, load_dataset, load_metric
+
+from transformers import AutoFeatureExtractor, pipeline
+
+
+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_regex = '[,?.!\-\;\:"“%‘”�—’…–]'  # noqa: W605 IMPORTANT: this should correspond to the chars that were ignored during training
+
+    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
+    feature_extractor = AutoFeatureExtractor.from_pretrained(args.model_id)
+    sampling_rate = feature_extractor.sampling_rate
+
+    # resample audio
+    dataset = dataset.cast_column("audio", Audio(sampling_rate=sampling_rate))
+
+    # load eval pipeline
+    asr = pipeline("automatic-speech-recognition", model=args.model_id)
+
+    # map function to decode audio
+    def map_to_pred(batch):
+        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)
+
+    # 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 5 seconds."
+    )
+    parser.add_argument(
+        "--stride_length_s", type=float, default=None, help="Stride of the audio chunks. Defaults to 1 second."
+    )
+    parser.add_argument(
+        "--log_outputs", action="store_true", help="If defined, write outputs to log file for analysis."
+    )
+    args = parser.parse_args()
+
+    main(args)

log_mozilla-foundation_common_voice_7_0_or_test_predictions.txt

@@ -0,0 +1,20 @@
+0
+ପ<s>ୋ<s>ର<s>ରେ<s> <s>ବ<s>ା<s>ଲ<s>ା<s> ଗ<s>ସ୍<s>ତ<s>ି<s> ଭ<s>ା<s>ଣ୍ଡ<s>ୀ<s> <s>ଗ<s>ୋ<s>ପ<s>ା<s>ଳ<s> ପ<s>ରୋ<s>ଠ<s>ା<s>ରୁ<s> <s>ଦ<s>େ<s>ଢ଼<s> କ<s>ୋ<s>ସ<s> <s>ଦ<s>ୁ<s>ର<s> ।
+1
+ଆ<s>ଜ<s>ି<s>ମ<s>ଳ<s>ତ<s>ୀ<s>ର<s> <s>ଆ<s>ଦ<s>ର<s> <s>ଦ<s>େ<s>ଖ<s>େ<s> <s>କ<s>ି<s>ଏ
+2
+ହ<s>ୁ<s>ଏ<s> <s>ତ<s> <s>ମ<s>ଙ୍<s>ଗ<s>ରା<s>ଜ<s>େ<s> ବ<s>ୁ<s>ଜ<s>ି<s>ଚ<s>ନ୍<s>ତ<s>ି<s> <s>ହ<s>ଲ<s>ଗ<s>ୁ<s> <s>ଚ<s>ି<s>ର<s>କ<s>ା<s>ଳ<s> ନ<s>େ<s>ମ<s>ନ୍<s>ତ<s>େ<s> ବ<s>ି<s>ଦ<s>ା<s>ୟ<s> <s>ଗ<s>ହ<s>ଣ<s> <s>କ<s>ଲ<s>ା<s> ।
+3
+ତ<s>ା<s>ହ<s>ା<s>ଲ<s>େ ଆ<s>ପ<s>ଣ<s> ଘ<s>ର<s> ପ<s>ା<s>ଣ<s>ି<s> ମ<s>ା<s>ଠ<s>ି<s>ଆା<s>ରୁ<s> <s>ତ<s> <s>ମ<s>ା<s>ଚ<s> ଭା<s>ହା<s>ର<s>ଥ<s>ାନ୍<s>ଦ<s>େ<s> ।
+4
+ତ<s>ହ<s>ି<s>ଁ <s>ଆ<s>ର<s> ଦ<s>ି<s>ନ<s> <s>ଠ<s>ା<s>ର<s>ୁ<s> <s>ଚ<s>ା<s>ର<s>ି<s>ଦ<s>ି<s>ନ<s> <s>ଜ<s>ା<s>ଏ<s> ଭ<s>ଗ<s>ୁ<s>ମ<s>ା<s>ନ<s>ଙ<s>କ<s> <s>ଗ<s>ା<s> ମ<s>ଦ<s>୍<s>ଆ<s>ର<s>େ<s> କ<s>େ<s>ହ<s>ି<s> <s>ଦ<s>େ<s>ଖ<s>ି<s> ନ<s>ା<s>ହ<s>ି<s>ଁ ।
+5
+ତ<s>େ<s>ବ<s>େ<s> ଆ<s>ଜ<s>ି<s> ଏ<s>ତ<s>େ<s> <s>ବ<s>ା<s>କ<s>ୁ<s>ଳ<s> <s>କ<s>ି<s> ପ<s>ା<s>ହ<s>ଁ
+6
+ବ<s>ା<s>ପ<s> <s>ଛ<s>େ<s>ଉ<s>ଉ<s>ଁ <s>ଡ<s>ଟ<s>ି<s> ଲ<s>ା<s>ଗ<s> <s>କ<s>ି<s>ଏ ଧ<s>ା<s>ଇଁ<s>ବ<s> ।
+7
+ଆ<s>ଉ<s> <s>ମ<s>ୁ<s>ଁ<s> ବ<s>ୋ<s>ଲ<s>ି<s> <s>ସ<s>ି<s>ନ<s>ା<s> <s>ତ<s>ୋ<s>ର<s> ମ<s>ା<s>ଣ<s> ପ<s>ା<s>ଙ୍<s>ଚ<s>ଟ<s>ଙ୍<s>କ<s>ା<s>ରେ<s> ନ<s>େ<s>ଲ<s>ି<s>
+8
+ବ<s>ା<s> <s>ହ<s>ାନ<s>ତ<s> <s>ପ<s>ା<s>କୁ<s>ଲ<s>ି<s>ର<s>େ<s> <s>ମ<s>ୁ<s>ହ<s>ଁ<s>ର<s>ୁ<s> <s>ପ<s>ୋ<s>ସ<s>ପ<s>ସ<s> ଜ<s>ା<s>ଳ<s> ପ<s>ୋ<s>ଛ<s>ି<s> ପ<s>କ<s>ା<s>ଉ<s>ଛ<s>ନ୍<s>ତ<s>ି<s> ।
+9
+ଏ<s>ଣ<s>ି<s>ଚ<s> <s>ପ<s>ି<s>ଚ୍<s>ୁ<s> ଲ<s>ୋ<s>କ<s> <s>ପ<s>ା<s>ଣ<s>ି<s> <s>ମ<s>ୁ<s>ଠ<s>ା<s>ଏ<s> ପ<s>ା<s>ଇ<s>ବ<s>ା<s>ର<s> <s>ବ<s>ା<s>ଟ<s> <s>କ<s>ର<s> ।

log_mozilla-foundation_common_voice_7_0_or_test_targets.txt

@@ -0,0 +1,20 @@
+0
+ମକ୍ରାମପୁରରେ ବାଲାଗସ୍ତି ଫାଣ୍ତି ଗୋପାଳପୁରଠାରୁ ଦେଢ଼କୋଶ ଦୂର ।
+1
+ଆଜି ମଳତୀର ଆଦର ଦେଖେ କିଏ
+2
+ହୁଏତ ମଙ୍ଗରାଜେ ବୁଝିଛନ୍ତି ଫଲ୍ଗୁ ଚିରକାଳ ନିମନ୍ତେ ବିଦାୟ ଗ୍ରହଣ କଲା ।
+3
+ତାହେଲେ ଆପଣଙ୍କ ଘର ପାଣିମାଠିଆରୁ ତ ମାଛ ବାହାରୁଥାନ୍ତେ
+4
+ତହିଁଆର ଦିନଠାରୁ ଚାରିଦିନ ଯାଏ ଭଗବାନକୁ ଗାଁ ମଧ୍ୟରେ କେହି ଦେଖିନାହିଁ ।
+5
+ତେବେ ଆଜି ଏତେ ବ୍ୟାକୁଳ କି ପାଇଁ
+6
+ବାପ ଛେଉଣ୍ଡଟି ଲାଗି କିଏ ଧାଇଁବ
+7
+ଆଉ ମୁଁ ବୋଲି ସିନା ତୋର ମାଣ ପାଞ୍ଚ ଟଙ୍କାରେ ନେଲି
+8
+ବାଁ ହାତ ପାପୁଲିରେ ମୁହଁରୁ ପୋଷ ପୋଷ ଝାଳ ପୋଛିପକାଉଅଛନ୍ତି ।
+9
+ଏଣିକି ପିତୃଲୋକ ପାଣି ମୁଠାଏ ପାଇବାର ବାଟ କର ।

mozilla-foundation_common_voice_7_0_or_test_eval_results.txt

@@ -0,0 +1,2 @@
+WER: 1.0921052631578947
+CER: 2.5547945205479454