Upload LM

added_tokens.json

@@ -0,0 +1 @@
+{"<s>": 55, "</s>": 56}

config.json

@@ -1,6 +1,6 @@
 {
-  "_name_or_path": "Harveenchadha/vakyansh-wav2vec2-urdu-urm-60",
-  "activation_dropout": 0.1,
+  "_name_or_path": "facebook/wav2vec2-xls-r-300m",
+  "activation_dropout": 0.0,
   "adapter_kernel_size": 3,
   "adapter_stride": 2,
   "add_adapter": false,
@@ -9,12 +9,11 @@
     "Wav2Vec2ForCTC"
   ],
   "attention_dropout": 0.1,
-  "bos_token": "<s>",
-  "bos_token_id": 0,
+  "bos_token_id": 1,
   "classifier_proj_size": 256,
-  "codevector_dim": 256,
+  "codevector_dim": 768,
   "contrastive_logits_temperature": 0.1,
-  "conv_bias": false,
+  "conv_bias": true,
   "conv_dim": [
     512,
     512,
@@ -45,43 +44,41 @@
   "ctc_loss_reduction": "mean",
   "ctc_zero_infinity": false,
   "diversity_loss_weight": 0.1,
-  "do_lower_case": false,
-  "do_stable_layer_norm": false,
-  "eos_token": "</s>",
+  "do_stable_layer_norm": true,
   "eos_token_id": 2,
   "feat_extract_activation": "gelu",
-  "feat_extract_norm": "group",
+  "feat_extract_dropout": 0.0,
+  "feat_extract_norm": "layer",
   "feat_proj_dropout": 0.0,
   "feat_quantizer_dropout": 0.0,
-  "final_dropout": 0.1,
+  "final_dropout": 0.0,
   "gradient_checkpointing": false,
   "hidden_act": "gelu",
   "hidden_dropout": 0.1,
-  "hidden_size": 768,
+  "hidden_size": 1024,
   "initializer_range": 0.02,
-  "intermediate_size": 3072,
+  "intermediate_size": 4096,
   "layer_norm_eps": 1e-05,
   "layerdrop": 0.0,
-  "mask_feature_length": 10,
+  "mask_feature_length": 64,
   "mask_feature_min_masks": 0,
-  "mask_feature_prob": 0.0,
+  "mask_feature_prob": 0.25,
   "mask_time_length": 10,
   "mask_time_min_masks": 2,
-  "mask_time_prob": 0.05,
+  "mask_time_prob": 0.75,
   "model_type": "wav2vec2",
   "num_adapter_layers": 3,
-  "num_attention_heads": 12,
+  "num_attention_heads": 16,
   "num_codevector_groups": 2,
   "num_codevectors_per_group": 320,
   "num_conv_pos_embedding_groups": 16,
   "num_conv_pos_embeddings": 128,
   "num_feat_extract_layers": 7,
-  "num_hidden_layers": 12,
+  "num_hidden_layers": 24,
   "num_negatives": 100,
-  "output_hidden_size": 768,
-  "pad_token": "[PAD]",
-  "pad_token_id": 1,
-  "proj_codevector_dim": 256,
+  "output_hidden_size": 1024,
+  "pad_token_id": 54,
+  "proj_codevector_dim": 768,
   "tdnn_dilation": [
     1,
     2,
@@ -104,10 +101,8 @@
     1
   ],
   "torch_dtype": "float32",
-  "transformers_version": "4.16.2",
-  "unk_token": "[UNK]",
+  "transformers_version": "4.16.0",
   "use_weighted_layer_sum": false,
-  "vocab_size": 48,
-  "word_delimiter_token": "|",
+  "vocab_size": 57,
   "xvector_output_dim": 512
 }

eval.py

@@ -1,153 +1,153 @@
-#!/usr/bin/env python3
-import argparse
-import re
-import unicodedata
-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())
-    text = unicodedata.normalize("NFKC", text)
-
-    # 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, device=0)
-
-    # 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)
+#!/usr/bin/env python3
+import argparse
+import re
+from typing import Dict
+
+import torch
+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())
+    text = re.sub("[،]", '', text)
+    text = re.sub("[؟]", '', text)
+    text = re.sub("['َ]", '', text)
+    text = re.sub("['ُ]", '', text)
+    text = re.sub("['ِ]", '', text)
+    text = re.sub("['ّ]", '', text)
+    text = re.sub("['ٔ]", '', text)
+    text = re.sub("['ٰ]", '', text)
+    # batch["sentence"] = re.sub("[ء]", '', batch["sentence"])
+    # batch["sentence"] = re.sub("[آ]", 'ا', batch["sentence"])
+    text = re.sub("[ۂ]", 'ہ', text)
+    text = re.sub("[ي]", "ی",text)
+    text = re.sub("[ؤ]", "و", text)
+    # batch["sentence"] = re.sub("[ئ]", 'ى', batch["sentence"])
+    text = re.sub("[ى]", 'ی', text)
+    text = re.sub("[۔]", '', text)
+
+    # 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
+    if args.device is None:
+        args.device = 0 if torch.cuda.is_available() else -1
+    asr = pipeline("automatic-speech-recognition", model=args.model_id, device=args.device)
+
+    # 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."
+    )
+    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)

preprocessor_config.json

@@ -3,7 +3,8 @@
   "feature_extractor_type": "Wav2Vec2FeatureExtractor",
   "feature_size": 1,
   "padding_side": "right",
-  "padding_value": 0,
-  "return_attention_mask": false,
+  "padding_value": 0.0,
+  "processor_class": "Wav2Vec2ProcessorWithLM",
+  "return_attention_mask": true,
   "sampling_rate": 16000
 }

special_tokens_map.json

@@ -1 +1 @@
-{"bos_token": "<s>", "eos_token": "</s>", "unk_token": "<unk>", "pad_token": "<pad>"}
+{"bos_token": "<s>", "eos_token": "</s>", "unk_token": "[UNK]", "pad_token": "[PAD]", "additional_special_tokens": [{"content": "<s>", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true}, {"content": "</s>", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true}, {"content": "<s>", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true}, {"content": "</s>", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true}]}

tokenizer_config.json

@@ -1 +1 @@
-{"unk_token": "<unk>", "bos_token": "<s>", "eos_token": "</s>", "pad_token": "<pad>", "do_lower_case": false, "word_delimiter_token": "|", "special_tokens_map_file": "/root/.cache/huggingface/transformers/898afae2faf3d20dd22149b80863bb5d448c67d9c9b821dd66753253a4d77c64.9d6cd81ef646692fb1c169a880161ea1cb95f49694f220aced9b704b457e51dd", "tokenizer_file": null, "name_or_path": "Harveenchadha/vakyansh-wav2vec2-urdu-urm-60", "tokenizer_class": "Wav2Vec2CTCTokenizer"}
+{"unk_token": "[UNK]", "bos_token": "<s>", "eos_token": "</s>", "pad_token": "[PAD]", "do_lower_case": false, "word_delimiter_token": "|", "special_tokens_map_file": null, "tokenizer_file": null, "name_or_path": "anuragshas/wav2vec2-large-xls-r-300m-ur-cv8", "tokenizer_class": "Wav2Vec2CTCTokenizer", "processor_class": "Wav2Vec2ProcessorWithLM"}

vocab.json

@@ -1 +1 @@
-{"<s>": 0, "<pad>": 1, "</s>": 2, "<unk>": 3, "|": 4, "ء": 5, "آ": 6, "ؤ": 7, "ئ": 8, "ا": 9, "ب": 10, "ت": 11, "ث": 12, "ج": 13, "ح": 14, "خ": 15, "د": 16, "ذ": 17, "ر": 18, "ز": 19, "س": 20, "ش": 21, "ص": 22, "ض": 23, "ط": 24, "ظ": 25, "ع": 26, "غ": 27, "ف": 28, "ق": 29, "ل": 30, "م": 31, "ن": 32, "و": 33, "ٹ": 34, "پ": 35, "چ": 36, "ڈ": 37, "ڑ": 38, "ژ": 39, "ک": 40, "گ": 41, "ں": 42, "ھ": 43, "ہ": 44, "ی": 45, "ے": 46, "ۓ": 47}
+{"ء": 1, "آ": 2, "ؤ": 3, "ئ": 4, "ا": 5, "ب": 6, "ت": 7, "ث": 8, "ج": 9, "ح": 10, "خ": 11, "د": 12, "ذ": 13, "ر": 14, "ز": 15, "س": 16, "ش": 17, "ص": 18, "ض": 19, "ط": 20, "ظ": 21, "ع": 22, "غ": 23, "ف": 24, "ق": 25, "ل": 26, "م": 27, "ن": 28, "و": 29, "ى": 30, "ي": 31, "ً": 32, "َ": 33, "ُ": 34, "ِ": 35, "ّ": 36, "ٔ": 37, "ٰ": 38, "ٹ": 39, "پ": 40, "چ": 41, "ڈ": 42, "ڑ": 43, "ژ": 44, "ک": 45, "گ": 46, "ں": 47, "ھ": 48, "ہ": 49, "ۂ": 50, "ی": 51, "ے": 52, "|": 0, "[UNK]": 53, "[PAD]": 54}