Add 5gram lang model

.gitattributes

@@ -26,3 +26,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zstandard filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
 wandb/run-20220201_153024-1w85vsuu/run-1w85vsuu.wandb filter=lfs diff=lfs merge=lfs -text
+language_model/unigrams.txt filter=lfs diff=lfs merge=lfs -text

add_kenlm.py

@@ -0,0 +1,34 @@
+import argparse
+from transformers import AutoProcessor
+from transformers import Wav2Vec2ProcessorWithLM
+from pyctcdecode import build_ctcdecoder
+
+
+def main(args):
+    processor = AutoProcessor.from_pretrained(args.model_name_or_path)
+    vocab_dict = processor.tokenizer.get_vocab()
+    sorted_vocab_dict = {
+        k.lower(): v for k, v in sorted(vocab_dict.items(), key=lambda item: item[1])
+    }
+    decoder = build_ctcdecoder(
+        labels=list(sorted_vocab_dict.keys()),
+        kenlm_model_path=args.kenlm_model_path,
+    )
+    processor_with_lm = Wav2Vec2ProcessorWithLM(
+        feature_extractor=processor.feature_extractor,
+        tokenizer=processor.tokenizer,
+        decoder=decoder,
+    )
+    processor_with_lm.save_pretrained(args.model_name_or_path)
+    print(f"Run: ~/bin/build_binary language_model/*.arpa language_model/5gram.bin -T $(pwd) && rm language_model/*.arpa")
+
+def parse_args():
+    parser = argparse.ArgumentParser()
+    parser.add_argument('--model_name_or_path', default="./", help='Model name or path. Defaults to ./')
+    parser.add_argument('--kenlm_model_path', required=True, help='Path to KenLM arpa file.')
+    args = parser.parse_args()
+    return args
+
+if __name__ == "__main__":
+    args = parse_args()
+    main(args)

alphabet.json

@@ -0,0 +1 @@
+{"labels": [" ", "a", "b", "c", "d", "e", "f", "g", "h", "i", "j", "k", "l", "m", "n", "o", "p", "q", "r", "s", "t", "u", "v", "w", "x", "y", "z", "\u00e5", "\u00e6", "\u00f8", "\u2047", "", "<s>", "</s>"], "is_bpe": false}

eval.py

@@ -6,7 +6,7 @@ from typing import Dict
 import torch
 from datasets import Audio, Dataset, load_dataset, load_metric
 
-from transformers import AutoFeatureExtractor, pipeline
+from transformers import AutoFeatureExtractor, AutoModel, AutoTokenizer, pipeline
 
 
 def log_results(result: Dataset, args: Dict[str, str]):
@@ -96,7 +96,12 @@ def main(args):
     # 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)
+    asr = pipeline("automatic-speech-recognition",
+        model=AutoModel.from_pretrained(args.model_id),
+        tokenizer=AutoTokenizer.from_pretrained(args.model_id),
+        feature_extractor=feature_extractor,
+        device=args.device
+    )
 
     # map function to decode audio
     def map_to_pred(batch):

language_model/5gram.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7b41c24c63f2f0585bea83666369593f3b3e6d047f327a90f36ebca2c35ef0ff
+size 4243671427

language_model/attrs.json

@@ -0,0 +1 @@
+{"alpha": 0.5, "beta": 1.5, "unk_score_offset": -10.0, "score_boundary": true}

language_model/unigrams.txt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ac3e71ca49838ca355df6fdcb8d89344a5a9bf9e1a76587cdf5df1367c19b9a9
+size 16759269

preprocessor_config.json

@@ -4,6 +4,7 @@
   "feature_size": 1,
   "padding_side": "right",
   "padding_value": 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]", "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}]}
+{"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}, {"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": null, "tokenizer_file": null, "name_or_path": "./", "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": "./", "tokenizer_class": "Wav2Vec2CTCTokenizer", "processor_class": "Wav2Vec2ProcessorWithLM"}