Wav2Vec2-Large-XLSR-53-Persian V2

Fine-tuned facebook/wav2vec2-large-xlsr-53 in Persian (Farsi) using Common Voice. 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:

Requirements

# 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 hazm

Prediction

import librosa
import torch
import torchaudio
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
from datasets import load_dataset

import numpy as np
import hazm
import re
import string

import IPython.display as ipd

_normalizer = hazm.Normalizer()

chars_to_ignore = [
    ",", "?", ".", "!", "-", ";", ":", '""', "%", "'", '"', "�",
    "#", "!", "؟", "?", "«", "»", "،", "(", ")", "؛", "'ٔ", "٬",'ٔ', ",", "?", 
    ".", "!", "-", ";", ":",'"',"“", "%", "‘", "”", "�", "–", "…", "_", "”", '“', '„',
    'ā', 'š',
    # "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", '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",
        
    # "ها": "  ها", "ئ": "ی",
        
    "a": " ای ", "b": " بی ", "c": " سی ", "d": " دی ", "e": " ایی ", "f": " اف ",
    "g": " جی ", "h": " اچ ", "i": " آی ", "j": " جی ", "k": " کی ", "l": " ال ",
    "m": " ام ", "n": " ان ", "o": " او ", "p": " پی ", "q": " کیو ", "r": " آر ",
    "s": " اس ", "t": " تی ", "u": " یو ", "v": " وی ", "w": " دبلیو ", "x": " اکس ",
    "y": " وای ", "z": " زد ",
    "\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

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)
    text = re.sub(" +", " ", text)
    text = text.strip() + " "

    batch["sentence"] = text
    return batch


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, 16_000)

    batch["speech"] = speech_array
    return batch


def predict(batch):
    features = processor(batch["speech"], sampling_rate=16_000, 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)[0]
    return batch


device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
processor = Wav2Vec2Processor.from_pretrained("m3hrdadfi/wav2vec2-large-xlsr-persian-v2")
model = Wav2Vec2ForCTC.from_pretrained("m3hrdadfi/wav2vec2-large-xlsr-persian-v2").to(device)

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]
    print("reference:", reference)
    print("predicted:", predicted)
    print('---')

Output:

reference: عجم زنده کردم بدین پارسی 
predicted: عجم زنده کردم بدین پارسی
---
reference: لباس هایم کی آماده خواهند شد 
predicted: لباس خایم کی آماده خواهند شد
---
reference: با مهان همنشین شدم 
predicted: با مهان همنشین شدم
---
reference: یکی از بهترین فیلم هایی بود که در این سال ها دیدم 
predicted: یکی از بهترین فیلمهایی بود که در این سالها دیدم
---
reference: اون خیلی بد ماساژ میده 
predicted: اون خیلی بد ماساژ میده
---
reference: هنوزم بزرگترین دستاورد دولت روحانی اینه که رییسی رییسجمهور نشد 
predicted: هنوزم بزرگترین دستآوردار دولت روانیاینه که ریسی ریسیومرو نشد
---
reference: واسه بدنسازی آماده ای 
predicted: واسه بعدنسافی آماده ای
---
reference: خدای من شماها سالمین 
predicted: خدای من شما ها سالمین
---
reference: بهشون ثابت میشه که دروغ نگفتم 
predicted: بهشون ثابت میشه که دروغ مگفتم
---
reference: آیا ممکن است یک پتو برای من بیاورید 
predicted: سف کمیتخ لظا
---
reference: نزدیک جلو 
predicted: رزیک جلو
---
reference: شایعه پراکن دربارهاش دروغ و شایعه می سازد 
predicted: شایه پراکن دربارهاش دروغ و شایعه می سازد
---
reference: وقتی نیاز است که یک چهره دوستانه بیابند 
predicted: وقتی نیاز است یک چهره دوستانه بیابند
---
reference: ممکنه رادیواکتیوی چیزی باشه 
predicted: ممکنه به آدیوتیوی چیزی باشه
---
reference: دهنتون رو ببندید 
predicted: دهن جن رو ببندید
---
reference: پاشیم بریم قند و شکر و روغنمون رو بگیریم تا تموم نشده 
predicted: پاشین بریم قند و شکر و روغنمون رو بگیریم تا تموم نشده
---
reference: اما قبل از تمام کردن بحث تاریخی باید ذکری هم از ناپیکس بکنیم 
predicted: اما قبل از تمام کردن بحث تاریخی باید ذکری هم از نایپکس بکنیم
---
reference: لطفا کپی امضا شده قرارداد را بازگردانید 
predicted: لطفا کپی امضال شده قرار داد را باز گردانید
---
reference: خیلی هم چیز مهمی نیست 
predicted: خیلی هم چیز مهمی نیست
---
reference: شایعه پراکن دربارهاش دروغ و شایعه می سازد 
predicted: شایه پراکن دربارهاش دروغ و شایعه می سازد
---

Evaluation

The model can be evaluated as follows on the Persian (Farsi) test data of Common Voice.

import librosa
import torch
import torchaudio
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
from datasets import load_dataset, load_metric

import numpy as np
import hazm
import re
import string

_normalizer = hazm.Normalizer()

chars_to_ignore = [
    ",", "?", ".", "!", "-", ";", ":", '""', "%", "'", '"', "�",
    "#", "!", "؟", "?", "«", "»", "،", "(", ")", "؛", "'ٔ", "٬",'ٔ', ",", "?", 
    ".", "!", "-", ";", ":",'"',"“", "%", "‘", "”", "�", "–", "…", "_", "”", '“', '„',
    'ā', 'š',
    # "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", '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",
        
    # "ها": "  ها", "ئ": "ی",
        
    "a": " ای ", "b": " بی ", "c": " سی ", "d": " دی ", "e": " ایی ", "f": " اف ",
    "g": " جی ", "h": " اچ ", "i": " آی ", "j": " جی ", "k": " کی ", "l": " ال ",
    "m": " ام ", "n": " ان ", "o": " او ", "p": " پی ", "q": " کیو ", "r": " آر ",
    "s": " اس ", "t": " تی ", "u": " یو ", "v": " وی ", "w": " دبلیو ", "x": " اکس ",
    "y": " وای ", "z": " زد ",
    "\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

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)
    text = re.sub(" +", " ", text)
    text = text.strip() + " "

    batch["sentence"] = text
    return batch


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, 16_000)

    batch["speech"] = speech_array
    return batch


def predict(batch):
    features = processor(batch["speech"], sampling_rate=16_000, 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)[0]
    return batch


device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
processor = Wav2Vec2Processor.from_pretrained("m3hrdadfi/wav2vec2-large-xlsr-persian-v2")
model = Wav2Vec2ForCTC.from_pretrained("m3hrdadfi/wav2vec2-large-xlsr-persian-v2").to(device)

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")
print("WER: {:.2f}".format(100 * wer.compute(predictions=result["predicted"], references=result["sentence"])))

Test Result:

• WER: 31.92%

Training

The Common Voice train, validation datasets were used for training.

You can see the training states here

The script used for training can be found here