# Audio Embedding with Vggish

*Author: Jael Gu*


## Desription

The audio embedding operator converts an input audio into a dense vector which can be used to represent the audio clip's semantics.
Each vector represents for an audio clip with a fixed length of around 0.9s.
This operator is built on top of [VGGish](https://github.com/tensorflow/models/tree/master/research/audioset/vggish) with Pytorch.
The model is a [VGG](https://arxiv.org/abs/1409.1556) variant pre-trained with a large scale of audio dataset [AudioSet](https://research.google.com/audioset).
As suggested, it is suitable to extract features at high level or warm up a larger model.

## Code Example

Generate embeddings for the audio "test.wav". 

 *Write the pipeline in simplified style*:

```python
from towhee import dc

dc.glob('test.wav')
  .audio_decode()
  .time_window(range=10)
  .audio_embedding.vggish()
  .show()
```
    | [-0.4931737, -0.40068552, -0.032327592, ...] shape=(10, 128) |

*Write a same pipeline with explicit inputs/outputs name specifications:*

```python
from towhee import dc

dc.glob['path']('test.wav')
  .audio_decode['path', 'audio']()
  .time_window['audio', 'frames'](range=10)
  .audio_embedding.vggish['frames', 'vecs']()
  .select('vecs')
  .to_vec()
```
    [array([[-0.4931737 , -0.40068552, -0.03232759, ..., -0.33428153,
          0.1333081 , -0.25221825],
        [-0.49023268, -0.40161428, -0.03255743, ..., -0.33395663,
          0.13261834, -0.25324696],
        [-0.4992406 , -0.39848825, -0.03186834, ..., -0.33684137,
          0.13326398, -0.25385314],
        ...,
        [-0.49047503, -0.40119144, -0.03144619, ..., -0.33282205,
          0.13334712, -0.2520305 ],
        [-0.48861542, -0.40097567, -0.03173053, ..., -0.33255234,
          0.13278192, -0.25157905],
        [-0.4886143 , -0.40098593, -0.03175077, ..., -0.3325425 ,
          0.13271847, -0.25159872]], dtype=float32)]

## Factory Constructor

Create the operator via the following factory method

***audio_embedding.vggish(weights_path=None, framework="pytorch")***

**Parameters:**

​   *weights_path: str*

​   The path to model weights. If None, it will load default model weights.

​   *framework: str*

​   The framework of model implementation.
Default value is "pytorch" since the model is implemented in Pytorch.

## Interface

An audio embedding operator generates vectors in numpy.ndarray given an audio file path or a [towhee audio](link/to/AudioFrame/api/doc).


**Parameters:**

​   *Union[str, towhee.types.Audio]*

​   The audio path or link in string.
Or audio input data in towhee audio frames.
The input data should represent for an audio longer than 0.9s.


**Returns**:

​   *numpy.ndarray*

​   Audio embeddings in shape (num_clips, 128).
Each embedding stands for features of an audio clip with length of 0.9s.