Wav2Vec2-Large-XLSR-53-Swedish

使用 NST Swedish Dictation 在瑞典语上对 Wav2Vec2-Large-XLSR-53 进行了微调。在使用此模型时，请确保语音输入采样率为16kHz。

使用方法

可以直接使用此模型（无需语言模型），方法如下：

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

test_dataset = load_dataset("common_voice", "sv-SE", split="test[:2%]").

processor = Wav2Vec2Processor.from_pretrained("KBLab/wav2vec2-large-xlsr-53-swedish")
model = Wav2Vec2ForCTC.from_pretrained("KBLab/wav2vec2-large-xlsr-53-swedish")

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])

评估

在 Common Voice 的瑞典语测试数据上，可以按照以下方法进行评估：

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

test_dataset = load_dataset("common_voice", "sv-SE", split="test")
wer = load_metric("wer")

processor = Wav2Vec2Processor.from_pretrained("KBLab/wav2vec2-large-xlsr-53-swedish")
model = Wav2Vec2ForCTC.from_pretrained("KBLab/wav2vec2-large-xlsr-53-swedish")
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"])))
print("CER: {:2f}".format(100 * wer.compute(predictions=[" ".join(list(entry)) for entry in result["pred_strings"]], references=[" ".join(list(entry)) for entry in result["sentence"]])))

词错误率（WER）：14.298610% 音素错误率（CER）：4.925294%

训练

首先，使用包含来自各个广播电台的1000小时瑞典语口语的语料库进一步预训练了XLSR模型50个epochs。其次，使用 NST Swedish Dictation 进行了微调，同时也使用了 Common Voice 。最后，仅使用了 Common Voice 数据集进行了最后的微调。使用了 Fairseq 脚本。