first commit

README.md

@@ -0,0 +1,123 @@
+---
+language: fr
+datasets:
+- common_voice
+tags:
+- audio
+- automatic-speech-recognition
+- speech
+- xlsr-fine-tuning
+license: apache-2.0
+model-index:
+- name: wav2vec2-large-xlsr-53-French_punctuation by Ilyes Rebai
+  results:
+  - task: 
+      name: Speech Recognition
+      type: automatic-speech-recognition
+    dataset:
+      name: Common Voice
+      args: fr
+    metrics:
+       - name: Test WER and CER on text and puctuation prediction
+         types: [wer, cer]
+         values:  [19.47%, 6.66%]
+       -name: Test WER and CER on text without punctuation
+         types: [wer, cer]
+         values:  [17.88%, 6.37%]
+
+---
+## Evaluation on Common Voice FR Test
+```python
+import re
+import torch
+import torchaudio
+from datasets import load_dataset, load_metric
+from transformers import (
+    Wav2Vec2ForCTC,
+    Wav2Vec2Processor,
+)
+
+
+
+model_name = "Ilyes/wav2vec2-large-xlsr-53-french_punctuation"
+
+
+model = Wav2Vec2ForCTC.from_pretrained(model_name).to('cuda')
+processor = Wav2Vec2Processor.from_pretrained(model_name)
+
+
+ds = load_dataset("common_voice", "fr", split="test")
+
+
+chars_to_ignore_regex = '[\;\:\"\“\%\‘\”\�\‘\’\’\’\‘\…\·\ǃ\«\‹\»\›“\”\\ʿ\ʾ\„\∞\\|\;\:\*\—\–\─\―\_\/\:\ː\;\=\«\»\→]'
+def normalize_text(text):
+    text = text.lower().strip()
+    text = re.sub('œ', 'oe', text)
+    text = re.sub('æ', 'ae', text)
+    text = re.sub("’|´|′|ʼ|‘|ʻ|`", "'", 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)
+    text = re.sub(chars_to_ignore_regex, '', text)
+    return text
+
+
+
+def map_to_array(batch):
+    speech, _ = torchaudio.load(batch["path"])
+    batch["speech"] = resampler.forward(speech.squeeze(0)).numpy()
+    batch["sampling_rate"] = resampler.new_freq
+    batch["sentence"] = normalize_text(batch["sentence"])
+    return batch
+
+ds = ds.map(map_to_array)
+
+resampler = torchaudio.transforms.Resample(48_000, 16_000)
+def map_to_pred(batch):
+    features = processor(batch["speech"], sampling_rate=batch["sampling_rate"][0], padding=True, return_tensors="pt")
+    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)
+    batch["target"] = batch["sentence"]
+    # remove duplicates
+    batch["target"] = re.sub('\.+', '.', batch["target"])
+    batch["target"] = re.sub('\?+', '?', batch["target"])
+    batch["target"] = re.sub('!+', '!', batch["target"])
+    batch["target"] = re.sub(',+', ',', batch["target"])
+    return batch
+
+result = ds.map(map_to_pred, batched=True, batch_size=16, remove_columns=list(ds.features.keys()))
+wer = load_metric("wer")
+print(wer.compute(predictions=result["predicted"], references=result["target"]))
+```
+## Some results
+
+| Reference  | Prediction |
+| ------------- | ------------- |
+| il vécut à new york et y enseigna une grande partie de sa vie. | il a vécu à new york et y enseigna une grande partie de sa vie. |
+| au classement par nations, l'allemagne est la tenante du titre. | au classement der nation l'allemagne est la tenante du titre.. |
+| voici un petit calcul pour fixer les idées. | voici un petit calcul pour fixer les idées. |
+| oh! tu dois être beau avec | oh! tu dois être beau avec. |
+| babochet vous le voulez? | baboche, vous le voulez? |
+| la commission est, par conséquent, défavorable à cet amendement. | la commission est, par conséquent, défavorable à cet amendement. |
+
+All the references and predictions of the test corpus are already available in this repository.
+
+## Results
+
+text + punctuation 
+
+WER=21.47% CER=7.21%
+
+
+text (without punctuation)
+
+WER=19.71% CER=6.91%
+

config.json

@@ -0,0 +1,76 @@
+{
+  "_name_or_path": "/workspace/output_models/fr/wav2vec2-large-xlsr-53/checkpoint-14700",
+  "activation_dropout": 0.055,
+  "apply_spec_augment": true,
+  "architectures": [
+    "Wav2Vec2ForCTC"
+  ],
+  "attention_dropout": 0.094,
+  "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": 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": 0.04,
+  "final_dropout": 0.0,
+  "gradient_checkpointing": true,
+  "hidden_act": "gelu",
+  "hidden_dropout": 0.047,
+  "hidden_size": 1024,
+  "initializer_range": 0.02,
+  "intermediate_size": 4096,
+  "layer_norm_eps": 1e-05,
+  "layerdrop": 0.041,
+  "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": 0.4,
+  "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": 48,
+  "transformers_version": "4.5.0.dev0",
+  "vocab_size": 49
+}

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
+}

pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9f0c763c2c455ca6fa21a4644e7e37991dee8d5e436d5dcb73e21d9acd932275
+size 1262134743

special_tokens_map.json

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

vocab.json

@@ -0,0 +1 @@
+{"a": 0, "e": 1, "i": 2, "o": 3, "u": 4, "y": 5, "b": 6, "c": 7, "d": 8, "f": 9, "g": 10, "h": 11, "j": 12, "k": 13, "l": 14, "m": 15, "n": 16, "p": 17, "q": 18, "r": 19, "s": 20, "t": 21, "v": 22, "w": 23, "x": 24, "z": 25, "à": 26, "â": 27, "ç": 28, "è": 29, "é": 30, "ê": 31, "ë": 32, "î": 33, "ï": 34, "ô": 35, "ù": 36, "û": 37, "ü": 38, "ÿ": 39, "|": 40, "'": 41, "-": 42, ".": 43, ",": 44, "!": 45, "?": 46, "<unk>": 47, "<pad>": 48}