mindspore.dataset.audio.ComputeDeltas
- class mindspore.dataset.audio.ComputeDeltas(win_length=5, pad_mode=BorderType.EDGE)[source]
Compute delta coefficients, also known as differential coefficients, of a spectrogram.
Delta coefficients help to understand the dynamics of the power spectrum. It can be computed using the following formula.
\[d_{t}=\frac{{\textstyle\sum_{n=1}^{N}}n(c_{t+n}-c_{t-n})}{2{\textstyle\sum_{n=1}^{N}}n^{2}}\]where \(d_{t}\) is the deltas at time \(t\) , \(c_{t}\) is the spectrogram coefficients at time \(t\) , \(N\) is \((\text{win_length} - 1) // 2\) .
- Parameters
win_length (int, optional) – The window length used for computing delta, must be no less than 3. Default:
5
.pad_mode (BorderType, optional) –
Mode parameter passed to padding, can be
BorderType.CONSTANT
,BorderType.EDGE
,BorderType.REFLECT
orBorderType.SYMMETRIC
. Default:BorderType.EDGE
.BorderType.CONSTANT
, pad with a constant value.BorderType.EDGE
, pad with the last value on the edge.BorderType.REFLECT
, reflect the value on the edge while omitting the last one. For example, pad [1, 2, 3, 4] with 2 elements on both sides will result in [3, 2, 1, 2, 3, 4, 3, 2].BorderType.SYMMETRIC
, reflect the value on the edge while repeating the last one. For example, pad [1, 2, 3, 4] with 2 elements on both sides will result in [2, 1, 1, 2, 3, 4, 4, 3].
- Raises
TypeError – If win_length is not of type int.
ValueError – If win_length is less than 3.
TypeError – If pad_mode is not of type
mindspore.dataset.audio.BorderType
.RuntimeError – If input tensor is not in shape of <…, freq, time>.
- Supported Platforms:
CPU
Examples
>>> import numpy as np >>> import mindspore.dataset as ds >>> import mindspore.dataset.audio as audio >>> >>> # Use the transform in dataset pipeline mode >>> waveform = np.random.random([5, 400 // 2 + 1, 30]) # 5 samples >>> numpy_slices_dataset = ds.NumpySlicesDataset(data=waveform, column_names=["audio"]) >>> transforms = [audio.ComputeDeltas(win_length=7, pad_mode=audio.BorderType.EDGE)] >>> numpy_slices_dataset = numpy_slices_dataset.map(operations=transforms, input_columns=["audio"]) >>> for item in numpy_slices_dataset.create_dict_iterator(num_epochs=1, output_numpy=True): ... print(item["audio"].shape, item["audio"].dtype) ... break (201, 30) float64 >>> >>> # Use the transform in eager mode >>> waveform = np.random.random([400 // 2 + 1, 30]) # 1 sample >>> output = audio.ComputeDeltas(win_length=7, pad_mode=audio.BorderType.EDGE)(waveform) >>> print(output.shape, output.dtype) (201, 30) float64
- Tutorial Examples: