Update readme for more info

README.md

@@ -37,10 +37,12 @@ model-index:
 
 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.
 
-## Usage
+## How To Use
 The model can be used directly (without a language model) as follows:
 
+### 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
@@ -49,176 +51,88 @@ The model can be used directly (without a language model) as follows:
 !pip install hazm
 ```
 
-```python
-import torch
-import torchaudio
-from datasets import load_dataset, load_metric
-from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
+### Preprocessing
 
-import librosa
-
-import pandas as pd
-import numpy as np
+```python
+# preprocessing the datasets.
+# normalizing the texts
 
 import hazm
-
-import random
-import os
-import string
-import six
 import re
+import string
 
-import IPython.display as ipd
 
-# Loading the datasets
-dataset = load_dataset("common_voice", "fa", split="test[:2%]")
+_normalizer = hazm.Normalizer()
+chars_to_ignore = [
+    ",", "?", ".", "!", "-", ";", ":", '""', "%", "'", '"', "�",
+    "#", "!", "؟", "?", "«", "»", "ء", "،", "(", ")", "؛", "'ٔ", "٬",'ٔ', ",", "?", 
+    ".", "!", "-", ";", ":",'"',"“", "%", "‘", "”", "�", "–", "…", "_", "”", '“', '„'
+]
+
+# In case of farsi
+# chars_to_ignore = chars_to_ignore + list(string.ascii_lowercase + string.digits)
+
+chars_to_mapping = {
+'ك': 'ک', 'دِ': 'د', 'بِ': 'ب', 'زِ': 'ز', 'ذِ': 'ذ', 'شِ': 'ش', 'سِ': 'س', 'ى': 'ی',
+'ي': 'ی', 'أ': 'ا', 'ؤ': 'و', "ے": "ی", "ۀ": "ه", "ﭘ": "پ", "ﮐ": "ک", "ﯽ": "ی",
+"ﺎ": "ا", "ﺑ": "ب", "ﺘ": "ت", "ﺧ": "خ", "ﺩ": "د", "ﺱ": "س", "ﻀ": "ض", "ﻌ": "ع",
+"ﻟ": "ل", "ﻡ": "م", "ﻢ": "م", "ﻪ": "ه", "ﻮ": "و", "ئ": "ی", 'ﺍ': "ا", 'ة': "ه",
+'ﯾ': "ی", 'ﯿ': "ی", 'ﺒ': "ب", 'ﺖ': "ت", 'ﺪ': "د", 'ﺮ': "ر", 'ﺴ': "س", 'ﺷ': "ش",
+'ﺸ': "ش", 'ﻋ': "ع", 'ﻤ': "م", 'ﻥ': "ن", 'ﻧ': "ن", 'ﻭ': "و", 'ﺭ': "ر", "ﮔ": "گ",
+"\u200c": " ", "\u200d": " ", "\u200e": " ", "\u200f": " ", "\ufeff": " ",
+}
+
+def multiple_replace(text, chars_to_mapping):
+    pattern = "|".join(map(re.escape, chars_to_mapping.keys()))
+    return re.sub(pattern, lambda m: chars_to_mapping[m.group()], str(text))
+
+def remove_special_characters(text, chars_to_ignore_regex):
+    text = re.sub(chars_to_ignore_regex, '', text).lower() + " "
+    return text
 
-device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
-processor = Wav2Vec2Processor.from_pretrained("m3hrdadfi/wav2vec2-large-xlsr-persian")
-model = Wav2Vec2ForCTC.from_pretrained("m3hrdadfi/wav2vec2-large-xlsr-persian").to(device)
+def normalizer(batch, chars_to_ignore, chars_to_mapping):
+    chars_to_ignore_regex = f"""[{"".join(chars_to_ignore)}]"""
+    text = batch["sentence"].lower().strip()
+    
+    text = _normalizer.normalize(text)
+    text = multiple_replace(text, chars_to_mapping)
+    text = remove_special_characters(text, chars_to_ignore_regex)
 
+    batch["sentence"] = text
+    return batch
+```
 
-# Preprocessing the datasets.
-# Normalizing the texts
-_normalizer = hazm.Normalizer()
-def multiple_replace(mapping, text):
-    pattern = "|".join(map(re.escape, mapping.keys()))
-    return re.sub(pattern, lambda m: mapping[m.group()], str(text))
-
-def convert_weirdos(input_str):
-    # character
-    mapping = {
-        'ك': 'ک',
-        'دِ': 'د',
-        'بِ': 'ب',
-        'زِ': 'ز',
-        'ذِ': 'ذ',
-        'شِ': 'ش',
-        'سِ': 'س',
-        'ى': 'ی',
-        'ي': 'ی',
-        'أ': 'ا',
-        'ؤ': 'و',
-        "ے": "ی",
-        "ۀ": "ه",
-        "ﭘ": "پ",
-        "ﮐ": "ک",
-        "ﯽ": "ی",
-        "ﺎ": "ا",
-        "ﺑ": "ب",
-        "ﺘ": "ت",
-        "ﺧ": "خ",
-        "ﺩ": "د",
-        "ﺱ": "س",
-        "ﻀ": "ض",
-        "ﻌ": "ع",
-        "ﻟ": "ل",
-        "ﻡ": "م",
-        "ﻢ": "م",
-        "ﻪ": "ه",
-        "ﻮ": "و",
-        "ئ": "ی",
-        'ﺍ': "ا",
-        'ة': "ه",
-        'ﯾ': "ی",
-        'ﯿ': "ی",
-        'ﺒ': "ب",
-        'ﺖ': "ت",
-        'ﺪ': "د",
-        'ﺮ': "ر",
-        'ﺴ': "س",
-        'ﺷ': "ش",
-        'ﺸ': "ش",
-        'ﻋ': "ع",
-        'ﻤ': "م",
-        'ﻥ': "ن",
-        'ﻧ': "ن",
-        'ﻭ': "و",
-        'ﺭ': "ر",
-        "ﮔ": "گ",
-    }
-
-    # notation
-    mapping.update(**{
-        "#": " ",
-        "!": " ",
-        "؟": " ",
-        "?": " ",
-        "«": " ",
-        "»": " ",
-        "ء": " ",
-        "،": " ",
-        "(": " ",
-        ")": " ",
-        "؛": " ",
-        "'ٔ": " ",
-        "٬": " ",
-        'ٔ': " ",
-        ",": " ",
-        "?": " ",
-        ".": " ",
-        "!": " ",
-        "-": " ",
-        ";": " ",
-        ":": " ",
-        '"': " ",
-        "“": " ",
-        "%": " ",
-        "‘": " ",
-        "”": " ",
-        "�": " ",
-        "–": " ",
-        "…": " ",
-        "_": " ",
-    })
-
-    return multiple_replace(mapping, input_str)
-
-
-PERSIAN_ALPHA = "\u0621-\u0628\u062A-\u063A\u0641-\u0642\u0644-\u0648\u064E-\u0651\u0655\u067E\u0686\u0698\u06A9\u06AF\u06BE\u06CC"  # noqa: E501
-PERSIAN_DIGIT = "\u06F0-\u06F9"
-
-COMMON_ARABIC_ALPHA = "\u0629\u0643\u0649-\u064B\u064D\u06D5"
-COMMON_ARABIC_DIGIT = "\u0660-\u0669"
-
-ZWNJ = "\u200c"
-
-ENGLISH = "a-z0-9\&"
-PERSIAN = PERSIAN_ALPHA + PERSIAN_DIGIT + COMMON_ARABIC_ALPHA + COMMON_ARABIC_DIGIT + ZWNJ
-
-
-def normalizer(text, min_ratio=1.1):
-    text = text.lower()
-    text = _normalizer.normalize(text)
-    text = text.replace("\u200c", " ")
-    text = text.replace("\u200d", " ")
-    text = text.replace("\u200e", " ")
-    text = text.replace("\u200f", " ")
-    text = text.replace("\ufeff", " ")
-    text = convert_weirdos(text)
+### Loading The Data
 
-    words = [word.replace("آ", "ا") if "آ" in word and not word.startswith("آ") else word for word in text.split()]
-    text = " ".join(words)
+```python
+from datasets import load_dataset
 
-#     if not text or not len(text) > 2:
-#         return None
+dataset = load_dataset("common_voice", "fa", split="test[:1%]")
+print(dataset)
+```
 
-#     en_text = re.sub(r"[^" + ENGLISH + "+]", " ", six.ensure_str(text))
-#     en_text = re.sub(r"\s+", " ", en_text)
-#     if len(en_text) > 1:
-#         return None
+**Output:**
+```text
+>>> 
+Dataset({
+    features: ['client_id', 'path', 'sentence', 'up_votes', 'down_votes', 'age', 'gender', 'accent', 'locale', 'segment'],
+    num_rows: 52
+})
+```
 
-    return text
+### Model
 
+```python
+import librosa
+import torch
+import torchaudio
+from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
+
+device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+processor = Wav2Vec2Processor.from_pretrained("m3hrdadfi/wav2vec2-large-xlsr-persian")
+model = Wav2Vec2ForCTC.from_pretrained("m3hrdadfi/wav2vec2-large-xlsr-persian").to(device)
 
-chars_to_ignore_regex = '[\,\?\.\!\-\;\:\"\“\%\‘\”\�]'
-def remove_special_characters(batch):
-    text = re.sub(chars_to_ignore_regex, '', batch["sentence"]).lower() + " "
-    text = normalizer(text)
-    batch["sentence"] = text
-    return batch
 
-# We need to read the aduio files as arrays
 def speech_file_to_array_fn(batch):
     speech_array, sampling_rate = torchaudio.load(batch["path"])
     speech_array = speech_array.squeeze().numpy()
@@ -227,6 +141,7 @@ def speech_file_to_array_fn(batch):
     batch["speech"] = speech_array
     return batch
 
+
 def predict(batch):
     features = processor(batch["speech"], sampling_rate=16_000, return_tensors="pt", padding=True)
 
@@ -240,15 +155,24 @@ def predict(batch):
 
     batch["predicted"] = processor.batch_decode(pred_ids)[0]
     return batch
-
-dataset = dataset.map(remove_special_characters)
-dataset = dataset.map(speech_file_to_array_fn, remove_columns=list(set(dataset.column_names) - set(['sentence', 'path'])))
-result = dataset.map(predict)
 ```
 
 ## Prediction
 
 ```python
+import IPython.display as ipd
+
+
+dataset = load_dataset("common_voice", "fa", split="test[:1%]")
+dataset = dataset.map(
+    normalizer, 
+    fn_kwargs={"chars_to_ignore": chars_to_ignore, "chars_to_mapping": chars_to_mapping},
+    remove_columns=list(set(dataset.column_names) - set(['sentence', 'path']))
+)
+
+dataset = dataset.map(speech_file_to_array_fn)
+result = dataset.map(predict)
+
 max_items = np.random.randint(0, len(result), 20).tolist()
 for i in max_items:
     reference, predicted =  result["sentence"][i], result["predicted"][i]
@@ -257,6 +181,7 @@ for i in max_items:
     print('---')
 ```
 
+**Output:**
 ```text
 reference: اطلاعات مسری است
 predicted: اطلاعات مسری است
@@ -321,9 +246,22 @@ predicted: من سفر کردم را دوست دارم
 
 ## Evaluation
 
+```bash
+mkdir cer
+wget -O cer/cer.py https://huggingface.co/ctl/wav2vec2-large-xlsr-cantonese/raw/main/cer.py
+```
+
 ```python
-!mkdir cer
-!wget -O cer/cer.py https://huggingface.co/ctl/wav2vec2-large-xlsr-cantonese/raw/main/cer.py
+from datasets import load_metric
+
+dataset = load_dataset("common_voice", "fa", split="test")
+dataset = dataset.map(
+    normalizer, 
+    fn_kwargs={"chars_to_ignore": chars_to_ignore, "chars_to_mapping": chars_to_mapping},
+    remove_columns=list(set(dataset.column_names) - set(['sentence', 'path']))
+)
+dataset = dataset.map(speech_file_to_array_fn)
+result = dataset.map(predict)
 
 wer = load_metric("wer")
 cer = load_metric("./cer")
@@ -332,9 +270,11 @@ print("WER: {:.2f}".format(100 * wer.compute(predictions=result["predicted"], re
 print("CER: {:.2f}".format(100 * cer.compute(predictions=result["predicted"], references=result["sentence"])))
 ```
 
-**Test Result**: 
-- WER: 32.09%
-- CER: 8.23%
+**Output:** 
+```text
+WER: 32.09%
+CER: 8.23%
+```
 
 
 ## Training