输入类型（Torch.cuda.floattensor）和重量类型（Torch.floattensor）应相同 - Torchaudio

Question

我不知道为什么将设备设置为“ cuda”后，我会收到错误”输入类型（torch.cuda.floattensor）和重量类型（Torch.floattensor）应该相同的“

每个Torchaudio对象都设置为

constructor中的cuda”： self.transformation = transformation.to（self.device）

在getItem方法中： signal = signal.to（self.device）

当设备硬编码为“ cpu”时，以下代码可以工作。

我正在分享整个代码，因为我不知道会出现什么问题。

import os
import torch
from torch.utils.data import Dataset
import pandas as pd
import torchaudio


class EmoDB(Dataset):

    def __init__(self, annotations_file, audio_dir, transformation,
                 target_sample_rate, num_samples, device):
        self.annotations = pd.read_csv(annotations_file)
        self.audio_dir = audio_dir
        self.device = device
        self.transformation = transformation.to(self.device)
        self.target_sample_rate = target_sample_rate
        self.num_samples = num_samples

    def __len__(self):
        return len(self.annotations)

    def __getitem__(self, index):
        audio_sample_path = self._get_audio_sample_path(index)
        label = self._get_audio_sample_label(index)
        signal, sr = torchaudio.load(audio_sample_path)
        signal = signal.to(self.device)
        # signal -> (num_channels, samples) -> (2, 16000) -> (1, 16000)
        signal = self._resample_if_necessary(signal, sr)
        signal = self._mix_down_if_necessary(signal)
        signal = self._cut_if_necessary(signal)
        signal = self._right_pad_if_necessary(signal)
        signal = self.transformation(signal)
        return signal, label

    def _cut_if_necessary(self, signal):
        # signal -> Tensor -> (1, num_samples)
        if signal.shape[1] > self.num_samples:
            signal = signal[:, :self.num_samples]
        return signal

    def _right_pad_if_necessary(self, signal):
        length_signal = signal.shape[1]
        if length_signal < self.num_samples:
            # [1, 1, 1] -> [1, 1, 1, 0, 0]
            num_missing_samples = self.num_samples - length_signal
            last_dim_padding = (0, num_missing_samples)
            # [1, 1, 1] -> [1,]
            signal = torch.nn.functional.pad(signal, last_dim_padding)
        return signal

    def _resample_if_necessary(self, signal, sr):
        if sr != self.target_sample_rate:
            resampler = torchaudio.transforms.Resample(sr, self.target_sample_rate)
            signal = resampler(signal)
        return signal

    def _mix_down_if_necessary(self, signal):
        if signal.shape[0] > 1:
            signal = torch.mean(signal, dim=0, keepdim=True)
        return signal

    def _get_audio_sample_path(self, index):
        path = os.path.join(self.audio_dir, self.annotations.iloc[
            index, 0])
        return path

    def _get_audio_sample_label(self, index):
        return self.annotations.iloc[index, 2]


if __name__ == "__main__":
    AUDIO_DIR = "./EmoDb_berlin_database/audio"
    ANNOTATIONS_FILE = "./EmoDb_berlin_database/metadata/EmoDB.csv"
    SAMPLE_RATE = 22050
    NUM_SAMPLES = 22050

    if torch.cuda.is_available():
        device = "cuda"
    else:
        device = "cpu"
    print(f"Using device {device}")

    mel_spectrogram = torchaudio.transforms.MelSpectrogram(
        sample_rate=SAMPLE_RATE,
        n_fft=1024,
        hop_length=512,
        n_mels=64
    )
    # hardcoding "device" as error
    # Input type (torch.cuda.FloatTensor) and weight type (torch.FloatTensor) should be the same
    # unsolved
    device = "cpu"
    emodb = EmoDB(ANNOTATIONS_FILE, AUDIO_DIR, mel_spectrogram,
                            SAMPLE_RATE, NUM_SAMPLES, device)
    print(f"There are {len(emodb)} samples in the dataset.")
    signal, label = emodb[0]

Answer 1

将重采样器发送到Cuda

resampler.to(self.device)