# TAUKADIAL: Speech-Based Cognitive Assessment in Chinese and English
[TOC]

[link](https://taukadial-luzs-69e3bf4b9878b99a6f03aea43776344580b77b9fe54725f4.gitlab.io)

:::info
## Deadlines:
* 20th February: registration deadline;
* 27th February: deadline for submission of results
* (Optional?) INTERSPEECH Paper Submission Deadline: 2 March 2024
* INTERSPEECH paper update deadline: 9 March 2024

## Tasks
1. Catergorization task
* healthy control speech v.s. MCI speech
2. Score Prediction task
* cognitive test score prediction

## Data
* The spontaneous speech samples corresponding to audio recordings of picture descriptions produced by cognitively normal subjects and patients with MCI
* speakers are Chinese and English

## Proposed process
*Image on taukadial site*
:::

## Current thoughts:
### Questions:
- [ ] Which task should we choose?
- [ ] Is the data participant-wise or audio-wise
- [ ] Should we extract interpretable features and develope a reasonable model? Or we focus on the performance of the model?
- [ ] What does **langauge-specific preprocessing** mean?
* Does it means the 'text features'? Or the specific auditory features (are they comparable between English and Chinese)?
- [ ] Should we calculate the errors in the description?
* How to determine the "right description"?

### How to extract features?
* Acoustic preprocessing:
* liborsa (pitch?)
* Speech rate? pause duration?
* Language-specific preprocessing:
-->how to speech to text?
:::info
(updated: if it is in the dementiabank, they might provide with the transcription. - 20240216 14:44)
:::
* BERT？ ( is the data enough?)
* Semantic unit, word counts, verb count.... number of sentences,...(see [Mueller et al., 2019](https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6198327/), [Vincze et al., 2022](https://direct.mit.edu/coli/article/48/1/119/108843/Linguistic-Parameters-of-Spontaneous-Speech-for))

### What models are used for prediction?
--> depends on the size of dataset.
--> [Tartarus](https://github.com/sergiooramas/tartarus)？
* Task 1:
* Traditional machine learning: Random forest, logistic regression, SVC
* Nerual network(?)-- How to deal with mulitmodal data
* Task 2: (under consrtuctions)