pub fn load_audio_with_resample( path: &str, device: &Device, target_sample_rate: Option<usize>, target_channels: Option<usize>, ) -> Result<Tensor>
return shape: (channel, audio_len)