wav2vec2 (ar, de, es, fr, fa, he, ro, ru, tr, uk, multi)

.gitattributes

@@ -33,3 +33,10 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+ar/wav2vec2-large-xlsr-53-arabic[[:space:]](elgeish)/validation_wer.png filter=lfs diff=lfs merge=lfs -text
+fa/wav2vec2-large-xlsr-persian-v3/sample2978.flac filter=lfs diff=lfs merge=lfs -text
+he/wav2vec2-xls-r-300m-lm-hebrew/language_model/unigrams.txt filter=lfs diff=lfs merge=lfs -text
+multi/wav2vec2-xlsr-53-espeak-cv-ft/img[[:space:]](issue[[:space:]]10).jpeg filter=lfs diff=lfs merge=lfs -text
+ru/wav2vec2-large-xlsr-53-russian/language_model/lm.binary filter=lfs diff=lfs merge=lfs -text
+ru/wav2vec2-large-xlsr-53-russian/language_model/unigrams.txt filter=lfs diff=lfs merge=lfs -text
+uk/w2v-xls-r-uk/language_model/lm.binary filter=lfs diff=lfs merge=lfs -text

ar/wav2vec2-large-xlsr-53-arabic (elgeish)/.gitattributes

@@ -0,0 +1,17 @@
+*.bin.* filter=lfs diff=lfs merge=lfs -text
+*.lfs.* filter=lfs diff=lfs merge=lfs -text
+*.bin filter=lfs diff=lfs merge=lfs -text
+*.h5 filter=lfs diff=lfs merge=lfs -text
+*.tflite filter=lfs diff=lfs merge=lfs -text
+*.tar.gz filter=lfs diff=lfs merge=lfs -text
+*.ot filter=lfs diff=lfs merge=lfs -text
+*.onnx filter=lfs diff=lfs merge=lfs -text
+*.arrow filter=lfs diff=lfs merge=lfs -text
+*.ftz filter=lfs diff=lfs merge=lfs -text
+*.joblib filter=lfs diff=lfs merge=lfs -text
+*.model filter=lfs diff=lfs merge=lfs -text
+*.msgpack filter=lfs diff=lfs merge=lfs -text
+*.pb filter=lfs diff=lfs merge=lfs -text
+*.pt filter=lfs diff=lfs merge=lfs -text
+*.pth filter=lfs diff=lfs merge=lfs -text
+*.msgpack filter=lfs diff=lfs merge=lfs -text

ar/wav2vec2-large-xlsr-53-arabic (elgeish)/README.md

@@ -0,0 +1,206 @@
+---
+language: ar
+datasets:
+- arabic_speech_corpus
+- mozilla-foundation/common_voice_6_1
+metrics:
+- wer
+tags:
+- audio
+- automatic-speech-recognition
+- speech
+- xlsr-fine-tuning-week
+- hf-asr-leaderboard
+license: apache-2.0
+model-index:
+- name: elgeish-wav2vec2-large-xlsr-53-arabic
+  results:
+  - task: 
+      name: Automatic Speech Recognition
+      type: automatic-speech-recognition
+    dataset:
+      name: Common Voice 6.1 (Arabic)
+      type: mozilla-foundation/common_voice_6_1
+      config: ar
+      split: test
+      args:
+        language: ar
+    metrics:
+       - name: Test WER
+         type: wer
+         value: 26.55
+       - name: Validation WER
+         type: wer
+         value: 23.39
+---
+
+# Wav2Vec2-Large-XLSR-53-Arabic
+
+Fine-tuned [facebook/wav2vec2-large-xlsr-53](https://huggingface.co/facebook/wav2vec2-large-xlsr-53)
+on Arabic using the `train` splits of [Common Voice](https://huggingface.co/datasets/common_voice)
+and [Arabic Speech Corpus](https://huggingface.co/datasets/arabic_speech_corpus).
+When using this model, make sure that your speech input is sampled at 16kHz.
+
+## Usage
+
+The model can be used directly (without a language model) as follows:
+
+```python
+import torch
+import torchaudio
+from datasets import load_dataset
+from lang_trans.arabic import buckwalter
+from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
+
+dataset = load_dataset("common_voice", "ar", split="test[:10]")
+resamplers = {  # all three sampling rates exist in test split
+    48000: torchaudio.transforms.Resample(48000, 16000),
+    44100: torchaudio.transforms.Resample(44100, 16000),
+    32000: torchaudio.transforms.Resample(32000, 16000),
+}
+
+def prepare_example(example):
+    speech, sampling_rate = torchaudio.load(example["path"])
+    example["speech"] = resamplers[sampling_rate](speech).squeeze().numpy()
+    return example
+
+dataset = dataset.map(prepare_example)
+processor = Wav2Vec2Processor.from_pretrained("elgeish/wav2vec2-large-xlsr-53-arabic")
+model = Wav2Vec2ForCTC.from_pretrained("elgeish/wav2vec2-large-xlsr-53-arabic").eval()
+
+def predict(batch):
+    inputs = processor(batch["speech"], sampling_rate=16000, return_tensors="pt", padding=True)
+    with torch.no_grad():
+        predicted = torch.argmax(model(inputs.input_values).logits, dim=-1)
+    predicted[predicted == -100] = processor.tokenizer.pad_token_id  # see fine-tuning script
+    batch["predicted"] = processor.tokenizer.batch_decode(predicted)
+    return batch
+
+dataset = dataset.map(predict, batched=True, batch_size=1, remove_columns=["speech"])
+
+for reference, predicted in zip(dataset["sentence"], dataset["predicted"]):
+    print("reference:", reference)
+    print("predicted:", buckwalter.untrans(predicted))
+    print("--")
+```
+
+Here's the output:
+
+```
+reference: ألديك قلم ؟
+predicted: هلديك قالر
+--
+reference: ليست هناك مسافة على هذه الأرض أبعد من يوم أمس.
+predicted: ليست نالك مسافة على هذه الأرض أبعد من يوم أمس
+--
+reference: إنك تكبر المشكلة.
+predicted: إنك تكبر المشكلة
+--
+reference: يرغب أن يلتقي بك.
+predicted: يرغب أن يلتقي بك
+--
+reference: إنهم لا يعرفون لماذا حتى.
+predicted: إنهم لا يعرفون لماذا حتى
+--
+reference: سيسعدني مساعدتك أي وقت تحب.
+predicted: سيسئدني مساعد سكرأي وقت تحب
+--
+reference: أَحَبُّ نظريّة علمية إليّ هي أن حلقات زحل مكونة بالكامل من الأمتعة المفقودة.
+predicted: أحب ناضريةً علمية إلي  هي أنحل قتزح المكونا بالكامل من الأمت عن المفقودة
+--
+reference: سأشتري له قلماً.
+predicted: سأشتري له قلما
+--
+reference: أين المشكلة ؟
+predicted: أين المشكل
+--
+reference: وَلِلَّهِ يَسْجُدُ مَا فِي السَّمَاوَاتِ وَمَا فِي الْأَرْضِ مِنْ دَابَّةٍ وَالْمَلَائِكَةُ وَهُمْ لَا يَسْتَكْبِرُونَ
+predicted: ولله يسجد ما في السماوات وما في الأرض من دابة والملائكة وهم لا يستكبرون
+--
+```
+
+## Evaluation
+
+The model can be evaluated as follows on the Arabic test data of Common Voice:
+
+```python
+import jiwer
+import torch
+import torchaudio
+from datasets import load_dataset
+from lang_trans.arabic import buckwalter
+from transformers import set_seed, Wav2Vec2ForCTC, Wav2Vec2Processor
+
+set_seed(42)
+test_split = load_dataset("common_voice", "ar", split="test")
+resamplers = {  # all three sampling rates exist in test split
+    48000: torchaudio.transforms.Resample(48000, 16000),
+    44100: torchaudio.transforms.Resample(44100, 16000),
+    32000: torchaudio.transforms.Resample(32000, 16000),
+}
+
+def prepare_example(example):
+    speech, sampling_rate = torchaudio.load(example["path"])
+    example["speech"] = resamplers[sampling_rate](speech).squeeze().numpy()
+    return example
+
+test_split = test_split.map(prepare_example)
+processor = Wav2Vec2Processor.from_pretrained("elgeish/wav2vec2-large-xlsr-53-arabic")
+model = Wav2Vec2ForCTC.from_pretrained("elgeish/wav2vec2-large-xlsr-53-arabic").to("cuda").eval()
+
+def predict(batch):
+    inputs = processor(batch["speech"], sampling_rate=16000, return_tensors="pt", padding=True)
+    with torch.no_grad():
+        predicted = torch.argmax(model(inputs.input_values.to("cuda")).logits, dim=-1)
+    predicted[predicted == -100] = processor.tokenizer.pad_token_id  # see fine-tuning script
+    batch["predicted"] = processor.batch_decode(predicted)
+    return batch
+
+test_split = test_split.map(predict, batched=True, batch_size=16, remove_columns=["speech"])
+transformation = jiwer.Compose([
+    # normalize some diacritics, remove punctuation, and replace Persian letters with Arabic ones
+    jiwer.SubstituteRegexes({
+        r'[auiFNKo\~_،؟»\?;:\-,\.؛«!"]': "", "\u06D6": "",
+        r"[\|\{]": "A", "p": "h", "ک": "k", "ی": "y"}),
+    # default transformation below
+    jiwer.RemoveMultipleSpaces(),
+    jiwer.Strip(),
+    jiwer.SentencesToListOfWords(),
+    jiwer.RemoveEmptyStrings(),
+])
+metrics = jiwer.compute_measures(
+    truth=[buckwalter.trans(s) for s in test_split["sentence"]],  # Buckwalter transliteration
+    hypothesis=test_split["predicted"],
+    truth_transform=transformation,
+    hypothesis_transform=transformation,
+)
+print(f"WER: {metrics['wer']:.2%}")
+```
+
+**Test Result**: 26.55%
+
+## Training
+
+For more details, see [Fine-Tuning with Arabic Speech Corpus](https://github.com/huggingface/transformers/tree/1c06240e1b3477728129bb58e7b6c7734bb5074e/examples/research_projects/wav2vec2#fine-tuning-with-arabic-speech-corpus).
+
+This model represents Arabic in a format called [Buckwalter transliteration](https://en.wikipedia.org/wiki/Buckwalter_transliteration).
+The Buckwalter format only includes ASCII characters, some of which are non-alpha (e.g., `">"` maps to `"أ"`).
+The [lang-trans](https://github.com/kariminf/lang-trans) package is used to convert (transliterate) Arabic abjad.
+
+[This script](https://github.com/huggingface/transformers/blob/1c06240e1b3477728129bb58e7b6c7734bb5074e/examples/research_projects/wav2vec2/finetune_large_xlsr_53_arabic_speech_corpus.sh)
+was used to first fine-tune [facebook/wav2vec2-large-xlsr-53](https://huggingface.co/facebook/wav2vec2-large-xlsr-53)
+on the `train` split of the [Arabic Speech Corpus](https://huggingface.co/datasets/arabic_speech_corpus) dataset;
+the `test` split was used for model selection; the resulting model at this point is saved as [elgeish/wav2vec2-large-xlsr-53-levantine-arabic](https://huggingface.co/elgeish/wav2vec2-large-xlsr-53-levantine-arabic).
+
+Training was then resumed using the `train` split of the [Common Voice](https://huggingface.co/datasets/common_voice) dataset;
+the `validation` split was used for model selection;
+training was stopped to meet the deadline of [Fine-Tune-XLSR Week](https://github.com/huggingface/transformers/blob/700229f8a4003c4f71f29275e0874b5ba58cd39d/examples/research_projects/wav2vec2/FINE_TUNE_XLSR_WAV2VEC2.md):
+this model is the checkpoint at 100k steps and a validation WER of **23.39%**.
+
+<img src="https://huggingface.co/elgeish/wav2vec2-large-xlsr-53-arabic/raw/main/validation_wer.png" alt="Validation WER" width="100%" />
+
+It's worth noting that validation WER is trending down, indicating the potential of further training (resuming the decaying learning rate at 7e-6).
+
+## Future Work
+One area to explore is using `attention_mask` in model input, which is recommended [here](https://huggingface.co/blog/fine-tune-xlsr-wav2vec2).
+Also, exploring data augmentation using datasets used to train models listed [here](https://paperswithcode.com/sota/speech-recognition-on-common-voice-arabic).

ar/wav2vec2-large-xlsr-53-arabic (elgeish)/config.json

@@ -0,0 +1,69 @@
+{
+  "_name_or_path": "elgeish/wav2vec2-large-xlsr-53-arabic",
+  "activation_dropout": 0.1,
+  "apply_spec_augment": true,
+  "architectures": [
+    "Wav2Vec2ForCTC"
+  ],
+  "attention_dropout": 0.1,
+  "bos_token_id": 1,
+  "conv_bias": true,
+  "conv_dim": [
+    512,
+    512,
+    512,
+    512,
+    512,
+    512,
+    512
+  ],
+  "conv_kernel": [
+    10,
+    3,
+    3,
+    3,
+    3,
+    2,
+    2
+  ],
+  "conv_stride": [
+    5,
+    2,
+    2,
+    2,
+    2,
+    2,
+    2
+  ],
+  "ctc_loss_reduction": "sum",
+  "ctc_zero_infinity": false,
+  "do_stable_layer_norm": true,
+  "eos_token_id": 2,
+  "feat_extract_activation": "gelu",
+  "feat_extract_dropout": 0.0,
+  "feat_extract_norm": "layer",
+  "feat_proj_dropout": 0.1,
+  "final_dropout": 0.1,
+  "gradient_checkpointing": false,
+  "hidden_act": "gelu",
+  "hidden_dropout": 0.1,
+  "hidden_dropout_prob": 0.1,
+  "hidden_size": 1024,
+  "initializer_range": 0.02,
+  "intermediate_size": 4096,
+  "layer_norm_eps": 1e-05,
+  "layerdrop": 0.1,
+  "mask_feature_length": 10,
+  "mask_feature_prob": 0.0,
+  "mask_time_length": 10,
+  "mask_time_prob": 0.05,
+  "model_type": "wav2vec2",
+  "num_attention_heads": 16,
+  "num_conv_pos_embedding_groups": 16,
+  "num_conv_pos_embeddings": 128,
+  "num_feat_extract_layers": 7,
+  "num_hidden_layers": 24,
+  "pad_token_id": 0,
+  "transformers_version": "4.4.0.dev0",
+  "vocab_size": 56
+}

ar/wav2vec2-large-xlsr-53-arabic (elgeish)/flax_model.msgpack

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:50f263c94e388555ccc40d13e8086de5222f7ca3a4aa20c4cad8232076e92fd4
+size 1261999872

ar/wav2vec2-large-xlsr-53-arabic (elgeish)/preprocessor_config.json

@@ -0,0 +1,8 @@
+{
+  "do_normalize": true,
+  "feature_size": 1,
+  "padding_side": "right",
+  "padding_value": 0.0,
+  "return_attention_mask": false,
+  "sampling_rate": 16000
+}

ar/wav2vec2-large-xlsr-53-arabic (elgeish)/pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fe4a13489e50820edbd709e85699dd87beeff5387e8c40d16860b10efb964547
+size 1262163415

ar/wav2vec2-large-xlsr-53-arabic (elgeish)/source.txt

@@ -0,0 +1 @@
+https://huggingface.co/elgeish/wav2vec2-large-xlsr-53-arabic

ar/wav2vec2-large-xlsr-53-arabic (elgeish)/special_tokens_map.json

@@ -0,0 +1 @@
+{"bos_token": "<s>", "eos_token": "</s>", "unk_token": "<unk>", "pad_token": "<pad>"}

ar/wav2vec2-large-xlsr-53-arabic (elgeish)/tokenizer_config.json

@@ -0,0 +1 @@
+{"unk_token": "<unk>", "bos_token": "<s>", "eos_token": "</s>", "pad_token": "<pad>", "do_lower_case": false, "word_delimiter_token": "/", "return_attention_mask": false, "do_normalize": true, "special_tokens_map_file": "special_tokens_map.json", "name_or_path": "elgeish/wav2vec2-large-xlsr-53-arabic"}

ar/wav2vec2-large-xlsr-53-arabic (elgeish)/vocab.json

@@ -0,0 +1 @@
+{"<pad>": 0, "<s>": 1, "</s>": 2, "<unk>": 3, "/": 4, "'": 5, "|": 6, ">": 7, "&": 8, "<": 9, "}": 10, "A": 11, "b": 12, "p": 13, "t": 14, "v": 15, "j": 16, "H": 17, "x": 18, "d": 19, "*": 20, "r": 21, "z": 22, "s": 23, "$": 24, "S": 25, "D": 26, "T": 27, "Z": 28, "E": 29, "g": 30, "_": 31, "f": 32, "q": 33, "k": 34, "l": 35, "m": 36, "n": 37, "h": 38, "w": 39, "Y": 40, "y": 41, "F": 42, "N": 43, "K": 44, "a": 45, "u": 46, "i": 47, "~": 48, "o": 49, "`": 50, "{": 51, "P": 52, "J": 53, "V": 54, "G": 55}

ar/wav2vec2-large-xlsr-53-arabic/.gitattributes

@@ -0,0 +1,17 @@
+*.bin.* filter=lfs diff=lfs merge=lfs -text
+*.lfs.* filter=lfs diff=lfs merge=lfs -text
+*.bin filter=lfs diff=lfs merge=lfs -text
+*.h5 filter=lfs diff=lfs merge=lfs -text
+*.tflite filter=lfs diff=lfs merge=lfs -text
+*.tar.gz filter=lfs diff=lfs merge=lfs -text
+*.ot filter=lfs diff=lfs merge=lfs -text
+*.onnx filter=lfs diff=lfs merge=lfs -text
+*.arrow filter=lfs diff=lfs merge=lfs -text
+*.ftz filter=lfs diff=lfs merge=lfs -text
+*.joblib filter=lfs diff=lfs merge=lfs -text
+*.model filter=lfs diff=lfs merge=lfs -text
+*.msgpack filter=lfs diff=lfs merge=lfs -text
+*.pb filter=lfs diff=lfs merge=lfs -text
+*.pt filter=lfs diff=lfs merge=lfs -text
+*.pth filter=lfs diff=lfs merge=lfs -text
+*.msgpack filter=lfs diff=lfs merge=lfs -text

ar/wav2vec2-large-xlsr-53-arabic/README.md

@@ -0,0 +1,200 @@
+---
+language: ar
+datasets:
+- common_voice
+- arabic_speech_corpus
+metrics:
+- wer
+- cer
+tags:
+- audio
+- automatic-speech-recognition
+- speech
+- xlsr-fine-tuning-week
+license: apache-2.0
+model-index:
+- name: XLSR Wav2Vec2 Arabic by Jonatas Grosman
+  results:
+  - task: 
+      name: Speech Recognition
+      type: automatic-speech-recognition
+    dataset:
+      name: Common Voice ar
+      type: common_voice
+      args: ar
+    metrics:
+       - name: Test WER
+         type: wer
+         value: 39.59
+       - name: Test CER
+         type: cer
+         value: 18.18
+---
+
+# Fine-tuned XLSR-53 large model for speech recognition in Arabic
+
+Fine-tuned [facebook/wav2vec2-large-xlsr-53](https://huggingface.co/facebook/wav2vec2-large-xlsr-53) on Arabic using the train and validation splits of [Common Voice 6.1](https://huggingface.co/datasets/common_voice) and [Arabic Speech Corpus](https://huggingface.co/datasets/arabic_speech_corpus).
+When using this model, make sure that your speech input is sampled at 16kHz.
+
+This model has been fine-tuned thanks to the GPU credits generously given by the [OVHcloud](https://www.ovhcloud.com/en/public-cloud/ai-training/) :)
+
+The script used for training can be found here: https://github.com/jonatasgrosman/wav2vec2-sprint
+
+## Usage
+
+The model can be used directly (without a language model) as follows...
+
+Using the [HuggingSound](https://github.com/jonatasgrosman/huggingsound) library:
+
+```python
+from huggingsound import SpeechRecognitionModel
+
+model = SpeechRecognitionModel("jonatasgrosman/wav2vec2-large-xlsr-53-arabic")
+audio_paths = ["/path/to/file.mp3", "/path/to/another_file.wav"]
+
+transcriptions = model.transcribe(audio_paths)
+```
+
+Writing your own inference script:
+
+```python
+import torch
+import librosa
+from datasets import load_dataset
+from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
+
+LANG_ID = "ar"
+MODEL_ID = "jonatasgrosman/wav2vec2-large-xlsr-53-arabic"
+SAMPLES = 10
+
+test_dataset = load_dataset("common_voice", LANG_ID, split=f"test[:{SAMPLES}]")
+
+processor = Wav2Vec2Processor.from_pretrained(MODEL_ID)
+model = Wav2Vec2ForCTC.from_pretrained(MODEL_ID)
+
+# Preprocessing the datasets.
+# We need to read the audio files as arrays
+def speech_file_to_array_fn(batch):
+    speech_array, sampling_rate = librosa.load(batch["path"], sr=16_000)
+    batch["speech"] = speech_array
+    batch["sentence"] = batch["sentence"].upper()
+    return batch
+
+test_dataset = test_dataset.map(speech_file_to_array_fn)
+inputs = processor(test_dataset["speech"], sampling_rate=16_000, return_tensors="pt", padding=True)
+
+with torch.no_grad():
+    logits = model(inputs.input_values, attention_mask=inputs.attention_mask).logits
+
+predicted_ids = torch.argmax(logits, dim=-1)
+predicted_sentences = processor.batch_decode(predicted_ids)
+
+for i, predicted_sentence in enumerate(predicted_sentences):
+    print("-" * 100)
+    print("Reference:", test_dataset[i]["sentence"])
+    print("Prediction:", predicted_sentence)
+```
+
+| Reference  | Prediction |
+| ------------- | ------------- |
+| ألديك قلم ؟ | ألديك قلم |
+| ليست هناك مسافة على هذه الأرض أبعد من يوم أمس. | ليست نالك مسافة على هذه الأرض أبعد من يوم الأمس  م |
+| إنك تكبر المشكلة. | إنك تكبر المشكلة |
+| يرغب أن يلتقي بك. | يرغب أن يلتقي بك |
+| إنهم لا يعرفون لماذا حتى. | إنهم لا يعرفون لماذا حتى |
+| سيسعدني مساعدتك أي وقت تحب. | سيسئدنيمساعدتك أي وقد تحب |
+| أَحَبُّ نظريّة علمية إليّ هي أن حلقات زحل مكونة بالكامل من الأمتعة المفقودة. | أحب نظرية علمية إلي  هي أن حل قتزح المكوينا بالكامل من الأمت عن المفقودة |
+| سأشتري له قلماً. | سأشتري له قلما |
+| أين المشكلة ؟ | أين المشكل |
+| وَلِلَّهِ يَسْجُدُ مَا فِي السَّمَاوَاتِ وَمَا فِي الْأَرْضِ مِنْ دَابَّةٍ وَالْمَلَائِكَةُ وَهُمْ لَا يَسْتَكْبِرُونَ | ولله يسجد ما في السماوات وما في الأرض من دابة والملائكة وهم لا يستكبرون |
+
+## Evaluation
+
+The model can be evaluated as follows on the Arabic test data of Common Voice.
+
+```python
+import torch
+import re
+import librosa
+from datasets import load_dataset, load_metric
+from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
+
+LANG_ID = "ar"
+MODEL_ID = "jonatasgrosman/wav2vec2-large-xlsr-53-arabic"
+DEVICE = "cuda"
+
+CHARS_TO_IGNORE = [",", "?", "¿", ".", "!", "¡", ";", "；", ":", '""', "%", '"', "�", "ʿ", "·", "჻", "~", "՞",
+                  "؟", "،", "।", "॥", "«", "»", "„", "“", "”", "「", "」", "‘", "’", "《", "》", "(", ")", "[", "]",
+                  "{", "}", "=", "`", "_", "+", "<", ">", "…", "–", "°", "´", "ʾ", "‹", "›", "©", "®", "—", "→", "。",
+                  "、", "﹂", "﹁", "‧", "～", "﹏", "，", "｛", "｝", "（", "）", "［", "］", "【", "】", "‥", "〽",
+                  "『", "』", "〝", "〟", "⟨", "⟩", "〜", "：", "！", "？", "♪", "؛", "/", "\\", "º", "−", "^", "'", "ʻ", "ˆ"]
+
+test_dataset = load_dataset("common_voice", LANG_ID, split="test")
+
+wer = load_metric("wer.py") # https://github.com/jonatasgrosman/wav2vec2-sprint/blob/main/wer.py
+cer = load_metric("cer.py") # https://github.com/jonatasgrosman/wav2vec2-sprint/blob/main/cer.py
+
+chars_to_ignore_regex = f"[{re.escape(''.join(CHARS_TO_IGNORE))}]"
+
+processor = Wav2Vec2Processor.from_pretrained(MODEL_ID)
+model = Wav2Vec2ForCTC.from_pretrained(MODEL_ID)
+model.to(DEVICE)
+
+# Preprocessing the datasets.
+# We need to read the audio files as arrays
+def speech_file_to_array_fn(batch):
+    with warnings.catch_warnings():
+        warnings.simplefilter("ignore")
+        speech_array, sampling_rate = librosa.load(batch["path"], sr=16_000)
+    batch["speech"] = speech_array
+    batch["sentence"] = re.sub(chars_to_ignore_regex, "", batch["sentence"]).upper()
+    return batch
+
+test_dataset = test_dataset.map(speech_file_to_array_fn)
+
+# Preprocessing the datasets.
+# We need to read the audio files as arrays
+def evaluate(batch):
+    inputs = processor(batch["speech"], sampling_rate=16_000, return_tensors="pt", padding=True)
+
+    with torch.no_grad():
+        logits = model(inputs.input_values.to(DEVICE), attention_mask=inputs.attention_mask.to(DEVICE)).logits
+
+    pred_ids = torch.argmax(logits, dim=-1)
+    batch["pred_strings"] = processor.batch_decode(pred_ids)
+    return batch
+
+result = test_dataset.map(evaluate, batched=True, batch_size=8)
+
+predictions = [x.upper() for x in result["pred_strings"]]
+references = [x.upper() for x in result["sentence"]]
+
+print(f"WER: {wer.compute(predictions=predictions, references=references, chunk_size=1000) * 100}")
+print(f"CER: {cer.compute(predictions=predictions, references=references, chunk_size=1000) * 100}")
+```
+
+**Test Result**:
+
+In the table below I report the Word Error Rate (WER) and the Character Error Rate (CER) of the model. I ran the evaluation script described above on other models as well (on 2021-05-14). Note that the table below may show different results from those already reported, this may have been caused due to some specificity of the other evaluation scripts used.
+
+| Model | WER | CER |
+| ------------- | ------------- | ------------- |
+| jonatasgrosman/wav2vec2-large-xlsr-53-arabic | **39.59%** | **18.18%** |
+| bakrianoo/sinai-voice-ar-stt | 45.30% | 21.84% |
+| othrif/wav2vec2-large-xlsr-arabic | 45.93% | 20.51% |
+| kmfoda/wav2vec2-large-xlsr-arabic | 54.14% | 26.07% |
+| mohammed/wav2vec2-large-xlsr-arabic | 56.11% | 26.79% |
+| anas/wav2vec2-large-xlsr-arabic | 62.02% | 27.09% |
+| elgeish/wav2vec2-large-xlsr-53-arabic | 100.00% | 100.56% |
+
+## Citation
+If you want to cite this model you can use this:
+
+```bibtex
+@misc{grosman2021xlsr53-large-arabic,
+  title={Fine-tuned {XLSR}-53 large model for speech recognition in {A}rabic},
+  author={Grosman, Jonatas},
+  howpublished={\url{https://huggingface.co/jonatasgrosman/wav2vec2-large-xlsr-53-arabic}},
+  year={2021}
+}
+```

ar/wav2vec2-large-xlsr-53-arabic/config.json

@@ -0,0 +1,76 @@
+{
+  "_name_or_path": "facebook/wav2vec2-large-xlsr-53",
+  "activation_dropout": 0.05,
+  "apply_spec_augment": true,
+  "architectures": [
+    "Wav2Vec2ForCTC"
+  ],
+  "attention_dropout": 0.1,
+  "bos_token_id": 1,
+  "conv_bias": true,
+  "conv_dim": [
+    512,
+    512,
+    512,
+    512,
+    512,
+    512,
+    512
+  ],
+  "conv_kernel": [
+    10,
+    3,
+    3,
+    3,
+    3,
+    2,
+    2
+  ],
+  "conv_stride": [
+    5,
+    2,
+    2,
+    2,
+    2,
+    2,
+    2
+  ],
+  "ctc_loss_reduction": "mean",
+  "ctc_zero_infinity": true,
+  "do_stable_layer_norm": true,
+  "eos_token_id": 2,
+  "feat_extract_activation": "gelu",
+  "feat_extract_dropout": 0.0,
+  "feat_extract_norm": "layer",
+  "feat_proj_dropout": 0.05,
+  "final_dropout": 0.0,
+  "gradient_checkpointing": true,
+  "hidden_act": "gelu",
+  "hidden_dropout": 0.05,
+  "hidden_size": 1024,
+  "initializer_range": 0.02,
+  "intermediate_size": 4096,
+  "layer_norm_eps": 1e-05,
+  "layerdrop": 0.05,
+  "mask_channel_length": 10,
+  "mask_channel_min_space": 1,
+  "mask_channel_other": 0.0,
+  "mask_channel_prob": 0.0,
+  "mask_channel_selection": "static",
+  "mask_feature_length": 10,
+  "mask_feature_prob": 0.0,
+  "mask_time_length": 10,
+  "mask_time_min_space": 1,
+  "mask_time_other": 0.0,
+  "mask_time_prob": 0.05,
+  "mask_time_selection": "static",
+  "model_type": "wav2vec2",
+  "num_attention_heads": 16,
+  "num_conv_pos_embedding_groups": 16,
+  "num_conv_pos_embeddings": 128,
+  "num_feat_extract_layers": 7,
+  "num_hidden_layers": 24,
+  "pad_token_id": 0,
+  "transformers_version": "4.5.0.dev0",
+  "vocab_size": 51
+}

ar/wav2vec2-large-xlsr-53-arabic/flax_model.msgpack

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b44a67c277854fbcd96179ee8bfedb9f03f3826efc2af35f8eb9b964fd0df2b1
+size 1261979372

ar/wav2vec2-large-xlsr-53-arabic/preprocessor_config.json

@@ -0,0 +1,8 @@
+{
+  "do_normalize": true,
+  "feature_size": 1,
+  "padding_side": "right",
+  "padding_value": 0.0,
+  "return_attention_mask": true,
+  "sampling_rate": 16000
+}

ar/wav2vec2-large-xlsr-53-arabic/pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a0b26f6d9d3edfde1784aef863c192a8cc1e438a23b45910ab648531ebe1857b
+size 1262142936

ar/wav2vec2-large-xlsr-53-arabic/source.txt

@@ -0,0 +1 @@
+https://huggingface.co/jonatasgrosman/wav2vec2-large-xlsr-53-arabic

ar/wav2vec2-large-xlsr-53-arabic/special_tokens_map.json

@@ -0,0 +1 @@
+{"bos_token": "<s>", "eos_token": "</s>", "unk_token": "<unk>", "pad_token": "<pad>"}

ar/wav2vec2-large-xlsr-53-arabic/vocab.json

@@ -0,0 +1 @@
+{"<pad>": 0, "<s>": 1, "</s>": 2, "<unk>": 3, "|": 4, "-": 5, "ء": 6, "آ": 7, "أ": 8, "ؤ": 9, "إ": 10, "ئ": 11, "ا": 12, "ب": 13, "ة": 14, "ت": 15, "ث": 16, "ج": 17, "ح": 18, "خ": 19, "د": 20, "ذ": 21, "ر": 22, "ز": 23, "س": 24, "ش": 25, "ص": 26, "ض": 27, "ط": 28, "ظ": 29, "ع": 30, "غ": 31, "ـ": 32, "ف": 33, "ق": 34, "ك": 35, "ل": 36, "م": 37, "ن": 38, "ه": 39, "و": 40, "ى": 41, "ي": 42, "ً": 43, "ٌ": 44, "ٍ": 45, "َ": 46, "ُ": 47, "ِ": 48, "ّ": 49, "ْ": 50}

de/wav2vec2-base-10k-voxpopuli-ft-de/.gitattributes

@@ -0,0 +1,17 @@
+*.bin.* filter=lfs diff=lfs merge=lfs -text
+*.lfs.* filter=lfs diff=lfs merge=lfs -text
+*.bin filter=lfs diff=lfs merge=lfs -text
+*.h5 filter=lfs diff=lfs merge=lfs -text
+*.tflite filter=lfs diff=lfs merge=lfs -text
+*.tar.gz filter=lfs diff=lfs merge=lfs -text
+*.ot filter=lfs diff=lfs merge=lfs -text
+*.onnx filter=lfs diff=lfs merge=lfs -text
+*.arrow filter=lfs diff=lfs merge=lfs -text
+*.ftz filter=lfs diff=lfs merge=lfs -text
+*.joblib filter=lfs diff=lfs merge=lfs -text
+*.model filter=lfs diff=lfs merge=lfs -text
+*.msgpack filter=lfs diff=lfs merge=lfs -text
+*.pb filter=lfs diff=lfs merge=lfs -text
+*.pt filter=lfs diff=lfs merge=lfs -text
+*.pth filter=lfs diff=lfs merge=lfs -text
+*.msgpack filter=lfs diff=lfs merge=lfs -text

de/wav2vec2-base-10k-voxpopuli-ft-de/README.md

@@ -0,0 +1,69 @@
+---
+language: de
+tags:
+- audio
+- automatic-speech-recognition
+- voxpopuli
+license: cc-by-nc-4.0
+---
+
+# Wav2Vec2-Base-VoxPopuli-Finetuned
+
+[Facebook's Wav2Vec2](https://ai.facebook.com/blog/wav2vec-20-learning-the-structure-of-speech-from-raw-audio/) base model pretrained on the 10K unlabeled subset of [VoxPopuli corpus](https://arxiv.org/abs/2101.00390) and fine-tuned on the transcribed data in de (refer to Table 1 of paper for more information).
+
+**Paper**: *[VoxPopuli: A Large-Scale Multilingual Speech Corpus for Representation
+Learning, Semi-Supervised Learning and Interpretation](https://arxiv.org/abs/2101.00390)*
+
+**Authors**: *Changhan Wang, Morgane Riviere, Ann Lee, Anne Wu, Chaitanya Talnikar, Daniel Haziza, Mary Williamson, Juan Pino, Emmanuel Dupoux* from *Facebook AI*
+
+See the official website for more information, [here](https://github.com/facebookresearch/voxpopuli/)
+
+
+# Usage for inference
+
+In the following it is shown how the model can be used in inference on a sample of the [Common Voice dataset](https://commonvoice.mozilla.org/en/datasets)
+
+```python
+#!/usr/bin/env python3
+from transformers import Wav2Vec2Processor, Wav2Vec2ForCTC
+from datasets import load_dataset
+import torchaudio
+import torch
+
+# resample audio
+
+# load model & processor
+model = Wav2Vec2ForCTC.from_pretrained("facebook/wav2vec2-base-10k-voxpopuli-ft-de")
+processor = Wav2Vec2Processor.from_pretrained("facebook/wav2vec2-base-10k-voxpopuli-ft-de")
+
+# load dataset
+ds = load_dataset("common_voice", "de", split="validation[:1%]")
+
+# common voice does not match target sampling rate
+common_voice_sample_rate = 48000
+target_sample_rate = 16000
+
+resampler = torchaudio.transforms.Resample(common_voice_sample_rate, target_sample_rate)
+
+
+# define mapping fn to read in sound file and resample
+def map_to_array(batch):
+    speech, _ = torchaudio.load(batch["path"])
+    speech = resampler(speech)
+    batch["speech"] = speech[0]
+    return batch
+
+
+# load all audio files
+ds = ds.map(map_to_array)
+
+# run inference on the first 5 data samples
+inputs = processor(ds[:5]["speech"], sampling_rate=target_sample_rate, return_tensors="pt", padding=True)
+
+# inference
+logits = model(**inputs).logits
+predicted_ids = torch.argmax(logits, axis=-1)
+
+print(processor.batch_decode(predicted_ids))
+```
+

de/wav2vec2-base-10k-voxpopuli-ft-de/config.json

@@ -0,0 +1,68 @@
+{
+  "activation_dropout": 0.1,
+  "apply_spec_augment": true,
+  "architectures": [
+    "Wav2Vec2ForCTC"
+  ],
+  "attention_dropout": 0.1,
+  "bos_token_id": 0,
+  "conv_bias": false,
+  "conv_dim": [
+    512,
+    512,
+    512,
+    512,
+    512,
+    512,
+    512
+  ],
+  "conv_kernel": [
+    10,
+    3,
+    3,
+    3,
+    3,
+    2,
+    2
+  ],
+  "conv_stride": [
+    5,
+    2,
+    2,
+    2,
+    2,
+    2,
+    2
+  ],
+  "ctc_loss_reduction": "sum",
+  "ctc_zero_infinity": false,
+  "do_stable_layer_norm": false,
+  "eos_token_id": 2,
+  "feat_extract_activation": "gelu",
+  "feat_extract_dropout": 0.0,
+  "feat_extract_norm": "group",
+  "feat_proj_dropout": 0.1,
+  "final_dropout": 0.1,
+  "gradient_checkpointing": false,
+  "hidden_act": "gelu",
+  "hidden_dropout": 0.1,
+  "hidden_dropout_prob": 0.1,
+  "hidden_size": 768,
+  "initializer_range": 0.02,
+  "intermediate_size": 3072,
+  "layer_norm_eps": 1e-05,
+  "layerdrop": 0.1,
+  "mask_feature_length": 10,
+  "mask_feature_prob": 0.0,
+  "mask_time_length": 10,
+  "mask_time_prob": 0.05,
+  "model_type": "wav2vec2",
+  "num_attention_heads": 12,
+  "num_conv_pos_embedding_groups": 16,
+  "num_conv_pos_embeddings": 128,
+  "num_feat_extract_layers": 7,
+  "num_hidden_layers": 12,
+  "pad_token_id": 1,
+  "transformers_version": "4.6.0.dev0",
+  "vocab_size": 36
+}

de/wav2vec2-base-10k-voxpopuli-ft-de/preprocessor_config.json

@@ -0,0 +1,9 @@
+{
+  "do_normalize": true,
+  "feature_extractor_type": "Wav2Vec2FeatureExtractor",
+  "feature_size": 1,
+  "padding_side": "right",
+  "padding_value": 0,
+  "return_attention_mask": false,
+  "sampling_rate": 16000
+}

de/wav2vec2-base-10k-voxpopuli-ft-de/pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6744100ba037593b22fac95738074e7cf6c1d9c94f0bfa0e76c3cf863b25741f
+size 377684844

de/wav2vec2-base-10k-voxpopuli-ft-de/source.txt

@@ -0,0 +1 @@
+https://huggingface.co/facebook/wav2vec2-base-10k-voxpopuli-ft-de

de/wav2vec2-base-10k-voxpopuli-ft-de/special_tokens_map.json

@@ -0,0 +1 @@
+{"bos_token": "<s>", "eos_token": "</s>", "unk_token": "<unk>", "pad_token": "<pad>"}

de/wav2vec2-base-10k-voxpopuli-ft-de/tokenizer_config.json

@@ -0,0 +1 @@
+{"unk_token": "<unk>", "bos_token": "<s>", "eos_token": "</s>", "pad_token": "<pad>", "do_lower_case": false, "word_delimiter_token": "|"}

de/wav2vec2-base-10k-voxpopuli-ft-de/vocab.json

@@ -0,0 +1 @@
+{"<pad>": 0, "<s>": 1, "</s>": 2, "<unk>": 3, "|": 4, "e": 5, "n": 6, "i": 7, "r": 8, "s": 9, "t": 10, "a": 11, "d": 12, "h": 13, "u": 14, "l": 15, "g": 16, "c": 17, "m": 18, "o": 19, "b": 20, "w": 21, "f": 22, "k": 23, "z": 24, "p": 25, "v": 26, "ü": 27, "ä": 28, "ö": 29, "j": 30, "ß": 31, "y": 32, "x": 33, "q": 34, "1": 35}

es/wav2vec2-large-es-voxpopuli/.gitattributes

@@ -0,0 +1,17 @@
+*.bin.* filter=lfs diff=lfs merge=lfs -text
+*.lfs.* filter=lfs diff=lfs merge=lfs -text
+*.bin filter=lfs diff=lfs merge=lfs -text
+*.h5 filter=lfs diff=lfs merge=lfs -text
+*.tflite filter=lfs diff=lfs merge=lfs -text
+*.tar.gz filter=lfs diff=lfs merge=lfs -text
+*.ot filter=lfs diff=lfs merge=lfs -text
+*.onnx filter=lfs diff=lfs merge=lfs -text
+*.arrow filter=lfs diff=lfs merge=lfs -text
+*.ftz filter=lfs diff=lfs merge=lfs -text
+*.joblib filter=lfs diff=lfs merge=lfs -text
+*.model filter=lfs diff=lfs merge=lfs -text
+*.msgpack filter=lfs diff=lfs merge=lfs -text
+*.pb filter=lfs diff=lfs merge=lfs -text
+*.pt filter=lfs diff=lfs merge=lfs -text
+*.pth filter=lfs diff=lfs merge=lfs -text
+*.msgpack filter=lfs diff=lfs merge=lfs -text

es/wav2vec2-large-es-voxpopuli/README.md

@@ -0,0 +1,24 @@
+---
+language: es
+tags:
+- audio
+- automatic-speech-recognition
+- voxpopuli
+license: cc-by-nc-4.0
+---
+
+# Wav2Vec2-Large-VoxPopuli
+
+[Facebook's Wav2Vec2](https://ai.facebook.com/blog/wav2vec-20-learning-the-structure-of-speech-from-raw-audio/) large model pretrained on the es unlabeled subset of [VoxPopuli corpus](https://arxiv.org/abs/2101.00390).
+
+**Paper**: *[VoxPopuli: A Large-Scale Multilingual Speech Corpus for Representation
+Learning, Semi-Supervised Learning and Interpretation](https://arxiv.org/abs/2101.00390)*
+
+**Authors**: *Changhan Wang, Morgane Riviere, Ann Lee, Anne Wu, Chaitanya Talnikar, Daniel Haziza, Mary Williamson, Juan Pino, Emmanuel Dupoux* from *Facebook AI*
+
+See the official website for more information, [here](https://github.com/facebookresearch/voxpopuli/)
+
+# Fine-Tuning
+
+Please refer to [this blog](https://huggingface.co/blog/fine-tune-xlsr-wav2vec2) on how to fine-tune this model on a specific language. Note that you should replace `"facebook/wav2vec2-large-xlsr-53"` with this checkpoint for fine-tuning.
+

es/wav2vec2-large-es-voxpopuli/config.json

@@ -0,0 +1,83 @@
+{
+  "activation_dropout": 0.0,
+  "apply_spec_augment": true,
+  "architectures": [
+    "Wav2Vec2ForPreTraining"
+  ],
+  "attention_dropout": 0.1,
+  "bos_token_id": 1,
+  "codevector_dim": 768,
+  "contrastive_logits_temperature": 0.1,
+  "conv_bias": true,
+  "conv_dim": [
+    512,
+    512,
+    512,
+    512,
+    512,
+    512,
+    512
+  ],
+  "conv_kernel": [
+    10,
+    3,
+    3,
+    3,
+    3,
+    2,
+    2
+  ],
+  "conv_stride": [
+    5,
+    2,
+    2,
+    2,
+    2,
+    2,
+    2
+  ],
+  "ctc_loss_reduction": "sum",
+  "ctc_zero_infinity": false,
+  "diversity_loss_weight": 0.1,
+  "do_stable_layer_norm": true,
+  "eos_token_id": 2,
+  "feat_extract_activation": "gelu",
+  "feat_extract_dropout": 0.0,
+  "feat_extract_norm": "layer",
+  "feat_proj_dropout": 0.1,
+  "feat_quantizer_dropout": 0.0,
+  "final_dropout": 0.0,
+  "gradient_checkpointing": false,
+  "hidden_act": "gelu",
+  "hidden_dropout": 0.1,
+  "hidden_size": 1024,
+  "initializer_range": 0.02,
+  "intermediate_size": 4096,
+  "layer_norm_eps": 1e-05,
+  "layerdrop": 0.1,
+  "mask_channel_length": 10,
+  "mask_channel_min_space": 1,
+  "mask_channel_other": 0.0,
+  "mask_channel_prob": 0.0,
+  "mask_channel_selection": "static",
+  "mask_feature_length": 10,
+  "mask_feature_prob": 0.0,
+  "mask_time_length": 10,
+  "mask_time_min_space": 1,
+  "mask_time_other": 0.0,
+  "mask_time_prob": 0.075,
+  "mask_time_selection": "static",
+  "model_type": "wav2vec2",
+  "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": 24,
+  "num_negatives": 100,
+  "pad_token_id": 0,
+  "proj_codevector_dim": 768,
+  "transformers_version": "4.7.0.dev0",
+  "vocab_size": 32
+}

es/wav2vec2-large-es-voxpopuli/flax_model.msgpack

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ad63febc6c296b8b3de75e39e739606d6bd239ab12c923b1dcdc176095dae2fd
+size 1269577963

es/wav2vec2-large-es-voxpopuli/preprocessor_config.json

@@ -0,0 +1,9 @@
+{
+  "do_normalize": true,
+  "feature_extractor_type": "Wav2Vec2FeatureExtractor",
+  "feature_size": 1,
+  "padding_side": "right",
+  "padding_value": 0,
+  "return_attention_mask": true,
+  "sampling_rate": 16000
+}

es/wav2vec2-large-es-voxpopuli/pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:92aa5fa25e1dd02a474cea12b2d958eacd128126f9ee8e8ce89249986e5da762
+size 1269737156

es/wav2vec2-large-es-voxpopuli/source.txt

@@ -0,0 +1 @@
+https://huggingface.co/facebook/wav2vec2-large-es-voxpopuli

fa/wav2vec2-large-xlsr-53-persian/.gitattributes

@@ -0,0 +1,17 @@
+*.bin.* filter=lfs diff=lfs merge=lfs -text
+*.lfs.* filter=lfs diff=lfs merge=lfs -text
+*.bin filter=lfs diff=lfs merge=lfs -text
+*.h5 filter=lfs diff=lfs merge=lfs -text
+*.tflite filter=lfs diff=lfs merge=lfs -text
+*.tar.gz filter=lfs diff=lfs merge=lfs -text
+*.ot filter=lfs diff=lfs merge=lfs -text
+*.onnx filter=lfs diff=lfs merge=lfs -text
+*.arrow filter=lfs diff=lfs merge=lfs -text
+*.ftz filter=lfs diff=lfs merge=lfs -text
+*.joblib filter=lfs diff=lfs merge=lfs -text
+*.model filter=lfs diff=lfs merge=lfs -text
+*.msgpack filter=lfs diff=lfs merge=lfs -text
+*.pb filter=lfs diff=lfs merge=lfs -text
+*.pt filter=lfs diff=lfs merge=lfs -text
+*.pth filter=lfs diff=lfs merge=lfs -text
+*.msgpack filter=lfs diff=lfs merge=lfs -text

fa/wav2vec2-large-xlsr-53-persian/README.md

@@ -0,0 +1,195 @@
+---
+language: fa
+datasets:
+- common_voice
+metrics:
+- wer
+- cer
+tags:
+- audio
+- automatic-speech-recognition
+- speech
+- xlsr-fine-tuning-week
+license: apache-2.0
+model-index:
+- name: XLSR Wav2Vec2 Persian by Jonatas Grosman
+  results:
+  - task: 
+      name: Speech Recognition
+      type: automatic-speech-recognition
+    dataset:
+      name: Common Voice fa
+      type: common_voice
+      args: fa
+    metrics:
+       - name: Test WER
+         type: wer
+         value: 30.12
+       - name: Test CER
+         type: cer
+         value: 7.37
+---
+
+# Fine-tuned XLSR-53 large model for speech recognition in Persian
+
+Fine-tuned [facebook/wav2vec2-large-xlsr-53](https://huggingface.co/facebook/wav2vec2-large-xlsr-53) on Persian using the train and validation splits of [Common Voice 6.1](https://huggingface.co/datasets/common_voice).
+When using this model, make sure that your speech input is sampled at 16kHz.
+
+This model has been fine-tuned thanks to the GPU credits generously given by the [OVHcloud](https://www.ovhcloud.com/en/public-cloud/ai-training/) :)
+
+The script used for training can be found here: https://github.com/jonatasgrosman/wav2vec2-sprint
+
+## Usage
+
+The model can be used directly (without a language model) as follows...
+
+Using the [HuggingSound](https://github.com/jonatasgrosman/huggingsound) library:
+
+```python
+from huggingsound import SpeechRecognitionModel
+
+model = SpeechRecognitionModel("jonatasgrosman/wav2vec2-large-xlsr-53-persian")
+audio_paths = ["/path/to/file.mp3", "/path/to/another_file.wav"]
+
+transcriptions = model.transcribe(audio_paths)
+```
+
+Writing your own inference script:
+
+```python
+import torch
+import librosa
+from datasets import load_dataset
+from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
+
+LANG_ID = "fa"
+MODEL_ID = "jonatasgrosman/wav2vec2-large-xlsr-53-persian"
+SAMPLES = 5
+
+test_dataset = load_dataset("common_voice", LANG_ID, split=f"test[:{SAMPLES}]")
+
+processor = Wav2Vec2Processor.from_pretrained(MODEL_ID)
+model = Wav2Vec2ForCTC.from_pretrained(MODEL_ID)
+
+# Preprocessing the datasets.
+# We need to read the audio files as arrays
+def speech_file_to_array_fn(batch):
+    speech_array, sampling_rate = librosa.load(batch["path"], sr=16_000)
+    batch["speech"] = speech_array
+    batch["sentence"] = batch["sentence"].upper()
+    return batch
+
+test_dataset = test_dataset.map(speech_file_to_array_fn)
+inputs = processor(test_dataset["speech"], sampling_rate=16_000, return_tensors="pt", padding=True)
+
+with torch.no_grad():
+    logits = model(inputs.input_values, attention_mask=inputs.attention_mask).logits
+
+predicted_ids = torch.argmax(logits, dim=-1)
+predicted_sentences = processor.batch_decode(predicted_ids)
+
+for i, predicted_sentence in enumerate(predicted_sentences):
+    print("-" * 100)
+    print("Reference:", test_dataset[i]["sentence"])
+    print("Prediction:", predicted_sentence)
+```
+
+| Reference  | Prediction |
+| ------------- | ------------- |
+| از مهمونداری کنار بکشم | از مهمانداری کنار بکشم |
+| برو از مهرداد بپرس. | برو از ماقدعاد به پرس |
+| خب ، تو چیكار می كنی؟ | خوب تو چیکار می کنی |
+| مسقط پایتخت عمان در عربی به معنای محل سقوط است | مسقط پایتخت عمان در عربی به بعنای محل سقوط است |
+| آه، نه اصلاُ! | اهنه اصلا |
+| توانست | توانست |
+| قصیده فن شعر میگوید ای دوستان | قصیده فن شعر میگوید ایدوستون |
+| دو استایل متفاوت دارین | دوبوست داریل و متفاوت بری |
+| دو روز قبل از کریسمس ؟ | اون مفتود پش پشش |
+| ساعت های کاری چیست؟ | این توری که موشیکل خب |
+
+## Evaluation
+
+The model can be evaluated as follows on the Persian test data of Common Voice.
+
+```python
+import torch
+import re
+import librosa
+from datasets import load_dataset, load_metric
+from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
+
+LANG_ID = "fa"
+MODEL_ID = "jonatasgrosman/wav2vec2-large-xlsr-53-persian"
+DEVICE = "cuda"
+
+CHARS_TO_IGNORE = [",", "?", "¿", ".", "!", "¡", ";", "；", ":", '""', "%", '"', "�", "ʿ", "·", "჻", "~", "՞",
+                   "؟", "،", "।", "॥", "«", "»", "„", "“", "”", "「", "」", "‘", "’", "《", "》", "(", ")", "[", "]",
+                   "{", "}", "=", "`", "_", "+", "<", ">", "…", "–", "°", "´", "ʾ", "‹", "›", "©", "®", "—", "→", "。",
+                   "、", "﹂", "﹁", "‧", "～", "﹏", "，", "｛", "｝", "（", "）", "［", "］", "【", "】", "‥", "〽",
+                   "『", "』", "〝", "〟", "⟨", "⟩", "〜", "：", "！", "？", "♪", "؛", "/", "\\", "º", "−", "^", "ʻ", "ˆ"]
+
+test_dataset = load_dataset("common_voice", LANG_ID, split="test")
+
+wer = load_metric("wer.py") # https://github.com/jonatasgrosman/wav2vec2-sprint/blob/main/wer.py
+cer = load_metric("cer.py") # https://github.com/jonatasgrosman/wav2vec2-sprint/blob/main/cer.py
+
+chars_to_ignore_regex = f"[{re.escape(''.join(CHARS_TO_IGNORE))}]"
+
+processor = Wav2Vec2Processor.from_pretrained(MODEL_ID)
+model = Wav2Vec2ForCTC.from_pretrained(MODEL_ID)
+model.to(DEVICE)
+
+# Preprocessing the datasets.
+# We need to read the audio files as arrays
+def speech_file_to_array_fn(batch):
+    with warnings.catch_warnings():
+        warnings.simplefilter("ignore")
+        speech_array, sampling_rate = librosa.load(batch["path"], sr=16_000)
+    batch["speech"] = speech_array
+    batch["sentence"] = re.sub(chars_to_ignore_regex, "", batch["sentence"]).upper()
+    return batch
+
+test_dataset = test_dataset.map(speech_file_to_array_fn)
+
+# Preprocessing the datasets.
+# We need to read the audio files as arrays
+def evaluate(batch):
+    inputs = processor(batch["speech"], sampling_rate=16_000, return_tensors="pt", padding=True)
+
+    with torch.no_grad():
+        logits = model(inputs.input_values.to(DEVICE), attention_mask=inputs.attention_mask.to(DEVICE)).logits
+
+    pred_ids = torch.argmax(logits, dim=-1)
+    batch["pred_strings"] = processor.batch_decode(pred_ids)
+    return batch
+
+result = test_dataset.map(evaluate, batched=True, batch_size=8)
+
+predictions = [x.upper() for x in result["pred_strings"]]
+references = [x.upper() for x in result["sentence"]]
+
+print(f"WER: {wer.compute(predictions=predictions, references=references, chunk_size=1000) * 100}")
+print(f"CER: {cer.compute(predictions=predictions, references=references, chunk_size=1000) * 100}")
+```
+
+**Test Result**:
+
+In the table below I report the Word Error Rate (WER) and the Character Error Rate (CER) of the model. I ran the evaluation script described above on other models as well (on 2021-04-22). Note that the table below may show different results from those already reported, this may have been caused due to some specificity of the other evaluation scripts used.
+
+| Model | WER | CER |
+| ------------- | ------------- | ------------- |
+| jonatasgrosman/wav2vec2-large-xlsr-53-persian | **30.12%** | **7.37%** |
+| m3hrdadfi/wav2vec2-large-xlsr-persian-v2 | 33.85% | 8.79% |
+| m3hrdadfi/wav2vec2-large-xlsr-persian | 34.37% | 8.98% |
+
+## Citation
+If you want to cite this model you can use this:
+
+```bibtex
+@misc{grosman2021xlsr53-large-persian,
+  title={Fine-tuned {XLSR}-53 large model for speech recognition in {P}ersian},
+  author={Grosman, Jonatas},
+  howpublished={\url{https://huggingface.co/jonatasgrosman/wav2vec2-large-xlsr-53-persian}},
+  year={2021}
+}
+```

fa/wav2vec2-large-xlsr-53-persian/config.json

@@ -0,0 +1,76 @@
+{
+  "_name_or_path": "facebook/wav2vec2-large-xlsr-53",
+  "activation_dropout": 0.05,
+  "apply_spec_augment": true,
+  "architectures": [
+    "Wav2Vec2ForCTC"
+  ],
+  "attention_dropout": 0.1,
+  "bos_token_id": 1,
+  "conv_bias": true,
+  "conv_dim": [
+    512,
+    512,
+    512,
+    512,
+    512,
+    512,
+    512
+  ],
+  "conv_kernel": [
+    10,
+    3,
+    3,
+    3,
+    3,
+    2,
+    2
+  ],
+  "conv_stride": [
+    5,
+    2,
+    2,
+    2,
+    2,
+    2,
+    2
+  ],
+  "ctc_loss_reduction": "mean",
+  "ctc_zero_infinity": true,
+  "do_stable_layer_norm": true,
+  "eos_token_id": 2,
+  "feat_extract_activation": "gelu",
+  "feat_extract_dropout": 0.0,
+  "feat_extract_norm": "layer",
+  "feat_proj_dropout": 0.05,
+  "final_dropout": 0.0,
+  "gradient_checkpointing": true,
+  "hidden_act": "gelu",
+  "hidden_dropout": 0.05,
+  "hidden_size": 1024,
+  "initializer_range": 0.02,
+  "intermediate_size": 4096,
+  "layer_norm_eps": 1e-05,
+  "layerdrop": 0.05,
+  "mask_channel_length": 10,
+  "mask_channel_min_space": 1,
+  "mask_channel_other": 0.0,
+  "mask_channel_prob": 0.0,
+  "mask_channel_selection": "static",
+  "mask_feature_length": 10,
+  "mask_feature_prob": 0.0,
+  "mask_time_length": 10,
+  "mask_time_min_space": 1,
+  "mask_time_other": 0.0,
+  "mask_time_prob": 0.05,
+  "mask_time_selection": "static",
+  "model_type": "wav2vec2",
+  "num_attention_heads": 16,
+  "num_conv_pos_embedding_groups": 16,
+  "num_conv_pos_embeddings": 128,
+  "num_feat_extract_layers": 7,
+  "num_hidden_layers": 24,
+  "pad_token_id": 0,
+  "transformers_version": "4.5.0.dev0",
+  "vocab_size": 67
+}

fa/wav2vec2-large-xlsr-53-persian/flax_model.msgpack

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f1c25fc4a3db03fb9610f8d954703e5b1497168dac2d4a5e67eaf1e400badb1f
+size 1262044974

fa/wav2vec2-large-xlsr-53-persian/issues.txt

@@ -0,0 +1,9 @@
+=============================================================================
+#3 WER 30% is wrong in real world/test
+=============================================================================
+
+[mosipvp] Feb 9, 2025
+
+Dear Creator, I think you are sharing an unreasonable result. I was test this in "Farsi" and the result is absolutely f***ing.
+I think you don't speak Farsi/Persian, so please use a native Farsi as your assistant/advisor.
+But thank you for sharing this model.

fa/wav2vec2-large-xlsr-53-persian/preprocessor_config.json

@@ -0,0 +1,8 @@
+{
+  "do_normalize": true,
+  "feature_size": 1,
+  "padding_side": "right",
+  "padding_value": 0.0,
+  "return_attention_mask": true,
+  "sampling_rate": 16000
+}

fa/wav2vec2-large-xlsr-53-persian/pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3b859c7f562a2cc3c6002c2eb5178b66777406c4fccf53f196ead46a4f6c4796
+size 1262208535

fa/wav2vec2-large-xlsr-53-persian/source.txt

@@ -0,0 +1 @@
+https://huggingface.co/jonatasgrosman/wav2vec2-large-xlsr-53-persian

fa/wav2vec2-large-xlsr-53-persian/special_tokens_map.json

@@ -0,0 +1 @@
+{"bos_token": "<s>", "eos_token": "</s>", "unk_token": "<unk>", "pad_token": "<pad>"}

fa/wav2vec2-large-xlsr-53-persian/vocab.json

@@ -0,0 +1 @@
+{"<pad>": 0, "<s>": 1, "</s>": 2, "<unk>": 3, "|": 4, "٬": 5, "و": 6, "ـ": 7, "ئ": 8, "ل": 9, "ج": 10, "ک": 11, "R": 12, "ِ": 13, "ع": 14, "َ": 15, "م": 16, "ض": 17, "-": 18, "I": 19, "F": 20, "ذ": 21, "ن": 22, "ژ": 23, "A": 24, "ش": 25, "ث": 26, "Y": 27, "د": 28, "ر": 29, "ّ": 30, "أ": 31, "ق": 32, "ب": 33, "ح": 34, "ظ": 35, "پ": 36, "ت": 37, "خ": 38, "غ": 39, "ط": 40, "ك": 41, "ي": 42, "E": 43, "Ā": 44, "؛": 45, "ی": 46, "چ": 47, "ه": 48, "M": 49, "ف": 50, "آ": 51, "ز": 52, "ص": 53, "س": 54, "گ": 55, "N": 56, "ُ": 57, "T": 58, "S": 59, "Š": 60, "ٔ": 61, "B": 62, "ء": 63, "ً": 64, "ا": 65, "ى": 66}

fa/wav2vec2-large-xlsr-persian-v3/.gitattributes

@@ -0,0 +1,17 @@
+*.bin.* filter=lfs diff=lfs merge=lfs -text
+*.lfs.* filter=lfs diff=lfs merge=lfs -text
+*.bin filter=lfs diff=lfs merge=lfs -text
+*.h5 filter=lfs diff=lfs merge=lfs -text
+*.tflite filter=lfs diff=lfs merge=lfs -text
+*.tar.gz filter=lfs diff=lfs merge=lfs -text
+*.ot filter=lfs diff=lfs merge=lfs -text
+*.onnx filter=lfs diff=lfs merge=lfs -text
+*.arrow filter=lfs diff=lfs merge=lfs -text
+*.ftz filter=lfs diff=lfs merge=lfs -text
+*.joblib filter=lfs diff=lfs merge=lfs -text
+*.model filter=lfs diff=lfs merge=lfs -text
+*.msgpack filter=lfs diff=lfs merge=lfs -text
+*.pb filter=lfs diff=lfs merge=lfs -text
+*.pt filter=lfs diff=lfs merge=lfs -text
+*.pth filter=lfs diff=lfs merge=lfs -text
+*tfevents* filter=lfs diff=lfs merge=lfs -text

fa/wav2vec2-large-xlsr-persian-v3/README.md

@@ -0,0 +1,236 @@
+---
+language: fa
+datasets:
+- common_voice
+tags:
+- audio
+- automatic-speech-recognition
+- speech
+- xlsr-fine-tuning-week
+widget:
+- example_title: Common Voice sample 1
+  src: https://huggingface.co/m3hrdadfi/wav2vec2-large-xlsr-persian-v3/resolve/main/sample1.flac
+- example_title: Common Voice sample 2978
+  src: https://huggingface.co/m3hrdadfi/wav2vec2-large-xlsr-persian-v3/resolve/main/sample2978.flac
+- example_title: Common Voice sample 5168
+  src: https://huggingface.co/m3hrdadfi/wav2vec2-large-xlsr-persian-v3/resolve/main/sample5168.flac
+model-index:
+- name: XLSR Wav2Vec2 Persian (Farsi) V3 by Mehrdad Farahani
+  results:
+  - task: 
+      name: Speech Recognition
+      type: automatic-speech-recognition
+    dataset:
+      name: Common Voice fa
+      type: common_voice
+      args: fa  
+    metrics:
+       - name: Test WER
+         type: wer
+         value: 10.36
+---
+
+# Wav2Vec2-Large-XLSR-53-Persian V3
+
+
+## Usage
+Fine-tuned [facebook/wav2vec2-large-xlsr-53](https://huggingface.co/facebook/wav2vec2-large-xlsr-53) in Persian (Farsi) using [Common Voice](https://huggingface.co/datasets/common_voice). When using this model, make sure that your speech input is sampled at 16kHz.
+
+
+**Requirements**
+```bash
+# requirement packages
+!pip install git+https://github.com/huggingface/datasets.git
+!pip install git+https://github.com/huggingface/transformers.git
+!pip install torchaudio
+!pip install librosa
+!pip install jiwer
+!pip install parsivar
+!pip install num2fawords
+```
+
+**Normalizer**
+```bash
+# Normalizer
+!wget -O normalizer.py https://huggingface.co/m3hrdadfi/"wav2vec2-large-xlsr-persian-v3/raw/main/dictionary.py
+!wget -O normalizer.py https://huggingface.co/m3hrdadfi/"wav2vec2-large-xlsr-persian-v3/raw/main/normalizer.py
+```
+
+**Downloading data**
+```bash
+wget https://voice-prod-bundler-ee1969a6ce8178826482b88e843c335139bd3fb4.s3.amazonaws.com/cv-corpus-6.1-2020-12-11/fa.tar.gz
+
+tar -xzf fa.tar.gz
+rm -rf fa.tar.gz
+```
+
+**Cleaning**
+```python
+from normalizer import normalizer
+
+def cleaning(text):
+    if not isinstance(text, str):
+        return None
+
+    return normalizer({"sentence": text}, return_dict=False)
+
+data_dir = "/content/cv-corpus-6.1-2020-12-11/fa"
+
+test = pd.read_csv(f"{data_dir}/test.tsv", sep="	")
+test["path"] = data_dir + "/clips/" + test["path"]
+print(f"Step 0: {len(test)}")
+
+test["status"] = test["path"].apply(lambda path: True if os.path.exists(path) else None)
+test = test.dropna(subset=["path"])
+test = test.drop("status", 1)
+print(f"Step 1: {len(test)}")
+
+test["sentence"] = test["sentence"].apply(lambda t: cleaning(t))
+test = test.dropna(subset=["sentence"])
+print(f"Step 2: {len(test)}")
+
+test = test.reset_index(drop=True)
+print(test.head())
+
+test = test[["path", "sentence"]]
+test.to_csv("/content/test.csv", sep="	", encoding="utf-8", index=False)
+```
+
+**Prediction**
+```python
+import numpy as np
+import pandas as pd
+
+import librosa
+import torch
+import torchaudio
+from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
+from datasets import load_dataset, load_metric
+
+import IPython.display as ipd
+
+model_name_or_path = "m3hrdadfi/wav2vec2-large-xlsr-persian-v3"
+device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+print(model_name_or_path, device)
+
+processor = Wav2Vec2Processor.from_pretrained(model_name_or_path)
+model = Wav2Vec2ForCTC.from_pretrained(model_name_or_path).to(device)
+
+
+def speech_file_to_array_fn(batch):
+    speech_array, sampling_rate = torchaudio.load(batch["path"])
+    speech_array = speech_array.squeeze().numpy()
+    speech_array = librosa.resample(np.asarray(speech_array), sampling_rate, processor.feature_extractor.sampling_rate)
+
+    batch["speech"] = speech_array
+    return batch
+
+
+def predict(batch):
+    features = processor(
+        batch["speech"], 
+        sampling_rate=processor.feature_extractor.sampling_rate, 
+        return_tensors="pt", 
+        padding=True
+    )
+
+    input_values = features.input_values.to(device)
+    attention_mask = features.attention_mask.to(device)
+
+    with torch.no_grad():
+        logits = model(input_values, attention_mask=attention_mask).logits 
+
+    pred_ids = torch.argmax(logits, dim=-1)
+
+    batch["predicted"] = processor.batch_decode(pred_ids)
+    return batch
+
+
+dataset = load_dataset("csv", data_files={"test": "/content/test.csv"}, delimiter="	")["test"]
+dataset = dataset.map(speech_file_to_array_fn)
+result = dataset.map(predict, batched=True, batch_size=4)
+```
+
+**WER Score**
+```python
+wer = load_metric("wer")
+print("WER: {:.2f}".format(100 * wer.compute(predictions=result["predicted"], references=result["sentence"])))
+```
+
+**Output**
+```python
+max_items = np.random.randint(0, len(result), 20).tolist()
+for i in max_items:
+    reference, predicted =  result["sentence"][i], result["predicted"][i]
+    print("reference:", reference)
+    print("predicted:", predicted)
+    print('---')
+```
+
+```text
+reference: ماجرا رو براش تعریف کردم او�� گفت مریم اگه میدونی پسر خوبیه خب چه اشکالی داره با‌هاش بیش‌تر اشنا بشو 
+predicted: ماجرا رو براش تعریف کردم اون گفت مریم اگه میدونی پسر خوبیه خب چه اشکالی داره با‌هاش بیش‌تر اشنا بشو
+---
+reference: بیا پایین تو اجازه نداری بری اون بالا 
+predicted: بیا پایین تو اجازه نداری بری اون بالا
+---
+reference: هر روز یک دو مداد کش می رفتتم تااین که تا پایان ترم از تمامی دوستانم مداد برداشته بودم 
+predicted: هر روز یک دو مداد کش می رفتم تااین که تا پایین ترم از تمامی دوستان و مداد برداشته بودم
+---
+reference: فکر میکنی آروم میشینه 
+predicted: فکر میکنی آروم میشینه
+---
+reference: هرکسی با گوشی هوشمند خود میتواند با کایلا متصل گردد در یک محدوده مکانی 
+predicted: هرکسی با گوشی هوشمند خود میتواند با کایلا متصل گردد در یک محدوده مکانی
+---
+reference: برو از مهرداد بپرس 
+predicted: برو از مهرداد بپرس
+---
+reference: می خواهم شما را با این قدم‌ها آشنا کنم 
+predicted: می خواهم شما را با این قدم‌ها آشنا کنم
+---
+reference: میدونم یه روز دوباره می تونم تو رو ببینم 
+predicted: میدونم یه روز دوباره می تونم تو رو ببینم
+---
+reference: بسیار خوب خواهد بود دعوت او را بپذیری 
+predicted: بسیار خوب خواهد بود دعوت او را بپذیری
+---
+reference: بهت بگن آشغالی خوبه 
+predicted: بهت بگن آشغالی خوبه
+---
+reference: چرا معاشرت با هم ایمانان ما را محفوظ نگه میدارد 
+predicted: چرا معاشرت با هم ایمانان آ را م حفوظ نگه میدارد
+---
+reference: بولیوی پس از گویان فقیر‌ترین کشور آمریکای جنوبی است 
+predicted: بولیوی پس از گویان فقیر‌ترین کشور آمریکای جنوبی است
+---
+reference: بعد از مدتی اینکار برایم عادی شد 
+predicted: بعد از مدتی اینکار برایم عادو شد
+---
+reference: به نظر اون هم همینطوره 
+predicted: به نظر اون هم همینطوره
+---
+reference: هیچ مایونز ی دارید 
+predicted: هیچ مایونز ی دارید
+---
+reference: هیچ یک از انان کاری به سنگ نداشتند 
+predicted: هیچ شک از انان کاری به سنگ نداشتند
+---
+reference: می خواهم کمی کتاب شعر ببینم 
+predicted: می خواهم کتاب شعر ببینم
+---
+reference: همین شوهر فهیمه مگه نمی گفتی فرمانده بوده کو 
+predicted: همین شوهر فهیمه بینامی گفتی فهمانده بود کو
+---
+reference: اون جا‌ها کسی رو نمیبینی که تو دستش کتاب نباشه 
+predicted: اون جا‌ها کسی رو نمیبینی که تو دستش کتاب نباشه
+---
+reference: زندان رفتن من در این سال‌های اخیر برام شانس بزرگی بود که معما و مشکل چندین سال‌هام را حل کرد 
+predicted: زندان رفتن من در این سال‌ها اخی براب شانس بزرگی بود که معما و مشکل چندین سال‌هام را حل کرد
+---
+```
+
+## Evaluation
+
+**Test Result:** 
+- WER: 10.36%

fa/wav2vec2-large-xlsr-persian-v3/config.json

@@ -0,0 +1,76 @@
+{
+  "_name_or_path": "facebook/wav2vec2-large-xlsr-53",
+  "activation_dropout": 0.09216,
+  "apply_spec_augment": true,
+  "architectures": [
+    "Wav2Vec2ForCTC"
+  ],
+  "attention_dropout": 0.05316,
+  "bos_token_id": 1,
+  "conv_bias": true,
+  "conv_dim": [
+    512,
+    512,
+    512,
+    512,
+    512,
+    512,
+    512
+  ],
+  "conv_kernel": [
+    10,
+    3,
+    3,
+    3,
+    3,
+    2,
+    2
+  ],
+  "conv_stride": [
+    5,
+    2,
+    2,
+    2,
+    2,
+    2,
+    2
+  ],
+  "ctc_loss_reduction": "mean",
+  "ctc_zero_infinity": true,
+  "do_stable_layer_norm": true,
+  "eos_token_id": 2,
+  "feat_extract_activation": "gelu",
+  "feat_extract_dropout": 0.0,
+  "feat_extract_norm": "layer",
+  "feat_proj_dropout": 0.01249,
+  "final_dropout": 0.0,
+  "gradient_checkpointing": true,
+  "hidden_act": "gelu",
+  "hidden_dropout": 0.01941,
+  "hidden_size": 1024,
+  "initializer_range": 0.02,
+  "intermediate_size": 4096,
+  "layer_norm_eps": 1e-05,
+  "layerdrop": 0.01377,
+  "mask_channel_length": 10,
+  "mask_channel_min_space": 1,
+  "mask_channel_other": 0.0,
+  "mask_channel_prob": 0.0,
+  "mask_channel_selection": "static",
+  "mask_feature_length": 10,
+  "mask_feature_prob": 0.0,
+  "mask_time_length": 10,
+  "mask_time_min_space": 1,
+  "mask_time_other": 0.0,
+  "mask_time_prob": 0.04529,
+  "mask_time_selection": "static",
+  "model_type": "wav2vec2",
+  "num_attention_heads": 16,
+  "num_conv_pos_embedding_groups": 16,
+  "num_conv_pos_embeddings": 128,
+  "num_feat_extract_layers": 7,
+  "num_hidden_layers": 24,
+  "pad_token_id": 0,
+  "transformers_version": "4.6.0.dev0",
+  "vocab_size": 40
+}