{
  "id": "1601.08188",
  "version": "v1",
  "published": "2016-01-29T16:48:07.000Z",
  "updated": "2016-01-29T16:48:07.000Z",
  "title": "Lipreading with Long Short-Term Memory",
  "authors": [
    "Michael Wand",
    "Jan Koutník",
    "Jürgen Schmidhuber"
  ],
  "comment": "Accepted for publication at ICASSP 2016",
  "categories": [
    "cs.CV",
    "cs.CL"
  ],
  "abstract": "Lipreading, i.e. speech recognition from visual-only recordings of a speaker's face, can be achieved with a processing pipeline based solely on neural networks, yielding significantly better accuracy than conventional methods. Feed-forward and recurrent neural network layers (namely Long Short-Term Memory; LSTM) are stacked to form a single structure which is trained by back-propagating error gradients through all the layers. The performance of such a stacked network was experimentally evaluated and compared to a standard Support Vector Machine classifier using conventional computer vision features (Eigenlips and Histograms of Oriented Gradients). The evaluation was performed on data from 19 speakers of the publicly available GRID corpus. With 51 different words to classify, we report a best word accuracy on held-out evaluation speakers of 79.6% using the end-to-end neural network-based solution (11.6% improvement over the best feature-based solution evaluated).",
  "revisions": [
    {
      "version": "v1",
      "updated": "2016-01-29T16:48:07.000Z"
    }
  ],
  "analyses": {
    "keywords": [
      "long short-term memory",
      "standard support vector machine classifier",
      "recurrent neural network layers",
      "conventional computer vision features",
      "held-out evaluation speakers"
    ],
    "note": {
      "typesetting": "TeX",
      "pages": 0,
      "language": "en",
      "license": "arXiv",
      "status": "editable",
      "adsabs": "2016arXiv160108188W"
    }
  }
}