AttributeError: 'Wav2Vec2FeatureExtractor' 对象没有属性 'decode'

Question

我正在尝试运行 wav2vec2 的简单示例并出现错误：

import torch
from transformers import Wav2Vec2ForCTC, Wav2Vec2FeatureExtractor
import librosa


model = Wav2Vec2ForCTC.from_pretrained("facebook/wav2vec2-large-xlsr-53")


feature_extractor = Wav2Vec2FeatureExtractor.from_pretrained("facebook/wav2vec2-large-xlsr-53")


SPEECH_FILE      = "/home/heb.wav"

waveform, sample_rate       = librosa.load(SPEECH_FILE, sr=16000)

features = feature_extractor(waveform, sampling_rate=sample_rate, return_tensors="pt")


with torch.no_grad():
    logits = model(features.input_values).logits

# Decode the predicted transcription from the logits
predicted_ids = torch.argmax(logits, dim=-1)
transcription = feature_extractor.decode(predicted_ids[0])

print(transcription)

我收到错误：

AttributeError: 'Wav2Vec2FeatureExtractor' object has no attribute 'decode'

如何解码 wav2vec2 输出？

（我使用的变形金刚版本：4.11.3）