Wav2vec2 Xlsr Nepali

🚀 Wav2Vec2-Large-XLSR-53-尼泊尔语模型

本模型基于 facebook/wav2vec2-large-xlsr-53，利用 Common Voice 和 OpenSLR ne 数据集在尼泊尔语上进行了微调。使用该模型时，请确保语音输入的采样率为 16kHz。

🔍 模型信息

🏋️‍ 评估结果

🚀 快速开始

本模型可直接使用（无需语言模型），示例代码如下：

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

!wget https://www.openslr.org/resources/43/ne_np_female.zip
!unzip ne_np_female.zip
!ls ne_np_female

colnames=['path','sentence'] 
df  = pd.read_csv('/content/ne_np_female/line_index.tsv',sep='\\t',header=None,names = colnames)
df['path'] = '/content/ne_np_female/wavs/'+df['path'] +'.wav'

train, test = train_test_split(df, test_size=0.1)

test.to_csv('/content/ne_np_female/line_index_test.csv')

test_dataset = load_dataset('csv', data_files='/content/ne_np_female/line_index_test.csv',split = 'train')

processor = Wav2Vec2Processor.from_pretrained("gagan3012/wav2vec2-xlsr-nepali")
model = Wav2Vec2ForCTC.from_pretrained("gagan3012/wav2vec2-xlsr-nepali") 

resampler = torchaudio.transforms.Resample(48_000, 16_000)

# Preprocessing the datasets.
# We need to read the aduio files as arrays
def speech_file_to_array_fn(batch):
	speech_array, sampling_rate = torchaudio.load(batch["path"])
	batch["speech"] = resampler(speech_array).squeeze().numpy()
	return batch

test_dataset = test_dataset.map(speech_file_to_array_fn)
inputs = processor(test_dataset["speech"][:2], 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)

print("Prediction:", processor.batch_decode(predicted_ids))
print("Reference:", test_dataset["sentence"][:2])

运行结果

Prediction: ['पारानाको ब्राजिली राज्यमा रहेको राजधानी', 'देवराज जोशी त्रिभुवन विश्वविद्यालयबाट शिक्षाशास्त्रमा स्नातक हुनुहुन्छ']
Reference: ['पारानाको ब्राजिली राज्यमा रहेको राजधानी', 'देवराज जोशी त्रिभुवन विश्वविद्यालयबाट शिक्षाशास्त्रमा स्नातक हुनुहुन्छ']

🧪 模型评估

可按以下方式在 Common Voice 的尼泊尔语测试数据上评估该模型：

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

!wget https://www.openslr.org/resources/43/ne_np_female.zip
!unzip ne_np_female.zip
!ls ne_np_female

colnames=['path','sentence'] 
df  = pd.read_csv('/content/ne_np_female/line_index.tsv',sep='\\t',header=None,names = colnames)
df['path'] = '/content/ne_np_female/wavs/'+df['path'] +'.wav'

train, test = train_test_split(df, test_size=0.1)

test.to_csv('/content/ne_np_female/line_index_test.csv')

test_dataset = load_dataset('csv', data_files='/content/ne_np_female/line_index_test.csv',split = 'train')
wer = load_metric("wer")

processor = Wav2Vec2Processor.from_pretrained("gagan3012/wav2vec2-xlsr-nepali")
model = Wav2Vec2ForCTC.from_pretrained("gagan3012/wav2vec2-xlsr-nepali") 
model.to("cuda")

chars_to_ignore_regex = '[\\,\\?\\.\\!\\-\\;\\:\\\"\\“]'  
resampler = torchaudio.transforms.Resample(48_000, 16_000)

# Preprocessing the datasets.
# We need to read the aduio files as arrays
def speech_file_to_array_fn(batch):
	batch["sentence"] = re.sub(chars_to_ignore_regex, '', batch["sentence"]).lower()
	speech_array, sampling_rate = torchaudio.load(batch["path"])
	batch["speech"] = resampler(speech_array).squeeze().numpy()
	return batch

test_dataset = test_dataset.map(speech_file_to_array_fn)

# Preprocessing the datasets.
# We need to read the aduio 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("cuda"), attention_mask=inputs.attention_mask.to("cuda")).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)

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

测试结果：5.97 %

🛠️ 模型训练

训练脚本可在 此处 找到。