Row convolution

As explained in Sutskever's List

Row convolution is a sequence-processing operation that applies learned convolutional transformations across rows or time steps of a representation.

Row convolution adds a narrow temporal convolution over unidirectional RNN outputs in Deep Speech 2. It approximates the benefits of bidirectional context while retaining low latency for streaming: unlike bidirectional RNNs, it does not require future inputs before making predictions. section 4.4.1

Figure 4.5 Left: DS2 architecture used to train on English and Mandarin speech. The authors vary the number of convolutional layers from 1 to 3 and the number of recurrent layers from 1 to 7. Right: structure of the RNN model from the first Deep Speech paper [45]. Used with permission granted by one of the authors (Awni Hannun).

Use the architecture diagram to locate row convolution within Deep Speech 2’s convolutional and recurrent stack. section 4.4.1

How Sutskever's List covers it

How deep it goes

Defines a narrow temporal convolution over unidirectional RNN outputs and contrasts its latency trade-off with bidirectional RNNs; implementation details are not given.

Examples

Contrasts past-only RNN predictions with bidirectional predictions that use both past and future context.

What it assumes

Recaps how unidirectional and bidirectional RNNs use temporal context, without establishing broader prerequisites.

What you can do after

Evaluate its role as a low-latency alternative to bidirectional RNN context in streaming speech recognition.

How it connects

Connects row convolution to unidirectional recurrent neural networks, bidirectional recurrent neural networks, and streaming speech recognition.

liveBookmanning.com

Read Sutskever's List on liveBook Get the book at manning.com

Related concepts

• Deep Speech 2used by · Deep Speech 2 is a large-scale neural speech-recognition system designed for end-to-end automatic speech recognition.

This explainer was generated by Manning's knowledge graph from the text of Sutskever's List. The section links point to the parts of the book it draws on.