Wav2Vec2-Large 语音说话者识别模型

模型描述

这是 S3PRL's Wav2Vec2 for the SUPERB Speaker Identification task 的转化版本。

基础模型是 wav2vec2-large-lv60 ，在预训练过程中使用了16kHz采样的语音音频。当使用该模型时，请确保输入的语音也是以16kHz采样。

更多信息请参考 SUPERB: Speech processing Universal PERformance Benchmark 。

任务和数据集描述

说话者识别（SI）将每个语音分段分类为其说话者身份，属于多类分类任务。训练和测试中使用了广泛采用的 VoxCeleb1 数据集。

有关原始模型的训练和评估说明，请参考 S3PRL downstream task README 。

使用示例

您可以通过音频分类流水线使用该模型：

from datasets import load_dataset
from transformers import pipeline

dataset = load_dataset("anton-l/superb_demo", "si", split="test")

classifier = pipeline("audio-classification", model="superb/wav2vec2-large-superb-sid")
labels = classifier(dataset[0]["file"], top_k=5)

或者直接使用该模型：

import torch
import librosa
from datasets import load_dataset
from transformers import Wav2Vec2ForSequenceClassification, Wav2Vec2FeatureExtractor

def map_to_array(example):
    speech, _ = librosa.load(example["file"], sr=16000, mono=True)
    example["speech"] = speech
    return example

# load a demo dataset and read audio files
dataset = load_dataset("anton-l/superb_demo", "si", split="test")
dataset = dataset.map(map_to_array)

model = Wav2Vec2ForSequenceClassification.from_pretrained("superb/wav2vec2-large-superb-sid")
feature_extractor = Wav2Vec2FeatureExtractor.from_pretrained("superb/wav2vec2-large-superb-sid")

# compute attention masks and normalize the waveform if needed
inputs = feature_extractor(dataset[:2]["speech"], sampling_rate=16000, padding=True, return_tensors="pt")

logits = model(**inputs).logits
predicted_ids = torch.argmax(logits, dim=-1)
labels = [model.config.id2label[_id] for _id in predicted_ids.tolist()]

评估结果

评估指标为准确率。

BibTeX 引用和文献信息

@article{yang2021superb,
  title={SUPERB: Speech processing Universal PERformance Benchmark},
  author={Yang, Shu-wen and Chi, Po-Han and Chuang, Yung-Sung and Lai, Cheng-I Jeff and Lakhotia, Kushal and Lin, Yist Y and Liu, Andy T and Shi, Jiatong and Chang, Xuankai and Lin, Guan-Ting and others},
  journal={arXiv preprint arXiv:2105.01051},
  year={2021}
}