Improved Physical Modeling of Speech Generation

[josh]

The computational approach may to modeling may also pay dividends for medical applications including speech therapy, interpretation of impaired speech, & providing machine speech support for impaired individuals.

Speech therapy – analyze which movements are particularly off & suggest word/sentence exercises to progressively build strength/fluency in creating those movements as well as hearing the direction of imprecision.

Interpretation of impaired speech – a better movement model can learn which elements of production are likely to be off and fit an overall interpretation for that PR context, as well as suggesting dialog to help confirm the diagnosis.

[josh]

As the Nature paper from Anumanchipalli et al. already suggests, thought patterns of neural activation can be linked to the sentences that the person wishes to convey as well as other elements of style/prosody.

[josh]

The contribution of the diaphragm to varying air flow pressure should also be modeled & sensed in some surrogate ways – e.g. chest muscle movements, water vapor leaving the mouth/nose, etc.

[josh]

We can also learn to make products that beautify individual vocal recordings while keeping the same person/id.

[josh]

Famous Voice Impersonators: How Do They Do It?

[Author]

Posts