Lecture 21 – DIVA model.docx

Lecture 21 – DIVA model

04.04.24

• 
• DIVA model explained
  • The speech sound map of the premotor cortex controls the articulator velocity and position maps (lower left) which, in turn, control the muscles of speech
  • The output of the cortex is fed to somatosensory and auditory error-map regions responsible for monitoring the accuracy of production based on the proposed plan produced at the premotor cortex
  • The error monitoring system serves as feedback control, which modifies the output over time during learning
  • Eventually the feedforward control has been modified sufficiently that feedback is no longer critically essential until a perturbation occurs in the system
    • That requires error correction (🡪 feedback loop)
  • Such a perturbation could be
    • Anesthesia from dental work
    • Or inserting a tube into the mouth
    • To a swollen lip or tongue
    • To motor disturbance arising from neuromuscular disease
    • Or simply an experiment where the auditory feedback is altered (phonetic experiments)
  • For normal speech the feedback loop is turned off (otherwise speech production would be too slow) 🡪 only activated when necessary (i.e. perturbation occurs)
  • General requirement for a speech model:
    • Feedback:
      • Auditory information (hearing your own voice and determine how exact targets have been reached)
      • Tactile and kinestheic feedback (how accurately your production was achieved in articulation (e.g. collision control or precision of closure/narrow channel for plosive vs fricative)
      • External sources (e.g. frown on someone’s face if they did not understand)
  • DIVA model contains mirror neurons model to match production to perception (a “model” of the desired speech sound is sent from the premotor cortex to the perceptual regions) 🡪 if mirror neuron “model” does not correspond to achieved acoustic output of the muscle model then muscle model is modified in the next trials 🡪 but this concept is still highly disputed
• DIVA model summary
  • The DIVA model established that learning takes place due to the feedback system of error correction
  • Learned accuracy carries us through task of talking (feedforward)
  • Failure to turn off the feedback leads to typical pathological speech pattern (e.g. stuttering)
  • Model can be used to drive speech synthesizers and thus help patients with pathological conditions (e.g. locked-in syndrome, speech motor disorders)
• DIVA model in applied sciences: stuttering
  • It takes 75 to 150 ms for auditory information to reach the cerebral cortex 🡪 that is too long to produce fluent running speech (e.g. an unstressed syllable is often only 100ms long)
  • So we have to disable the feedback loop for faster speech
    • 🡪 that is the reason why we speak slower when learning new languages (learning process)
    • 🡪dysfluency very similar to stuttering can be achieved when we do not disable the feedback part of the model and increase the speech rate -> in other words, in stuttering the feedback part of the model cannot be turned off, causing interference between the continuous auditory feedback and the motor planning for the following phonemes
• DIVA model in applied sciences: Locked-in syndrome
  • Sensors have been implanted in the model-appropriate reginos of a young patients with locked-in syndrome (following a stroke after a traumatic brain injury)
  • The researchers first identified the area of the atient’s brain that was active when the patient imagined himself speaking 🡪 these regions are likely to be involved in the active production of speech
  • They then implemented a “neural prosthesis” that could read activity in these regions
  • These activations run a speech synthesizer based on the specific phonemes that the patient is thinking of producing
  • This DIVA-human combination can now successfully and consistently produce three vowels