fix stft detect to support torchaudio.transforms.MelSpectrogram

[futz12]

we can directly export MelSpectrogram right now

import torchaudio
from torchaudio.transforms import MelSpectrogram
import torch

# Load an audio file
waveform = torch.randn(1, 1, 16000)
waveform2 = torch.randn(1, 1, 16000 + 160 * 5)
sample_rate = 16000

# Create a MelSpectrogram transform
mel_transform = MelSpectrogram(
    sample_rate=sample_rate,
    n_fft=1024,
    win_length=1024,
    hop_length=160,
    n_mels=128,
    f_min=30,
    f_max=8000,
)
# Apply the transform to the waveform
mel_spectrogram = mel_transform(waveform)
print(mel_spectrogram.shape)  # Output: (channels, n_mels, time)

import pnnx

pnnx.export(mel_transform, "mel_transform.pnnx", waveform, inputs2=waveform2)

(dev) PS C:\Users\wuyex\Documents\Project\ncnn\tools\pnnx\cmake-build-relwithdebinfo\src> python .\mel_transform_ncnn.py
tensor([[[[ 19.7871,  32.1392,  48.4337,  ...,  18.9732, 102.2494, 169.9629],
          [  4.8176,  17.9500,  36.9528,  ...,  64.0518, 131.9300, 158.6522],
          [  1.8964,   9.1304,  18.8681,  ...,  74.7491,  58.5998,  50.5858],
          ...,
          [335.0807, 376.1324, 419.6162,  ..., 246.3479, 321.4625, 366.1712],
          [136.8933, 144.3150, 168.2224,  ..., 342.1469, 265.8184, 225.4766],
          [396.7309, 368.3260, 291.3547,  ..., 229.3970, 219.3196, 220.0465]]]])
(dev) PS C:\Users\wuyex\Documents\Project\ncnn\tools\pnnx\cmake-build-relwithdebinfo\src> python .\mel_transform_pnnx.py

tensor([[[[ 19.7937,  32.1496,  48.4496,  ...,  18.9794, 102.2833, 170.0189],
          [  4.8202,  17.9608,  36.9756,  ...,  64.0931, 132.0132, 158.7513],
          [  1.8974,   9.1336,  18.8738,  ...,  74.7806,  58.6124,  50.5914],
          ...,
          [335.1889, 376.2388, 419.7093,  ..., 246.4327, 321.5598, 366.2824],
          [136.9423, 144.3594, 168.2723,  ..., 342.2699, 265.9131, 225.5431],
          [396.8740, 368.4949, 291.4909,  ..., 229.4855, 219.4119, 220.1355]]]])