Deep Learning for

Speech Accents in Games

Problem Formulation

  • We investigate how selective uses and distributions of voiced-over linguistic accents in video game characters may cast light on underlying language ideologies and, specifically, on accent attitudinal evaluations.
  • We attempt to analyze the language used in a database of video game recordings, paying specific attention to accents and all they might entail including:
      • The accents used by characters and what they signify.
      • The number of diverse accents used.
      • Whether or not a fantasy accent was created and why.
      • What accents reveal about someone's character.


  • We first train a deep neural network on data from the Speech Accent Database and then run the video game accents through the trained classifier.
  • The classifier network is convolutional and takes a spectrogram of accented speech as the input.

Sample spectrogram of a Spanish accent