Connected speech refers to natural, continuous spoken language produced in conversation or narrative, rather than isolated words or sentences. It provides rich acoustic and linguistic information on lexical retrieval, syntax, fluency, and speech rate, making it a sensitive measure of language impairment in PPA.

The administration and data analysis of the connected speech samples constitutes the research the Aim 2: Identify bilingualism factors associated with differential patterns of language impairment in Hispanics with PPA using metrics derived from connected speech.

Overview of the processes and this guide

 In connected speech, there are different processes, our main communication is completed though R01_CS_Data_Processing | Multilingual Aphasia and Dementia Research Lab | Microsoft Teams and in the https://cloud.wikis.utexas.edu/wiki/spaces/MADRWiki/pages/482182128

Detailed methodology

Open

Source: AoA_2025_Grasso_Santos

Training videos

Specific training videos of these processes. CS_Connected SpeechR01Protocol_Training Videos

Data structure overview

Connected speech overview structure, that explains how the files are saved in Box:

• Here are the Connected Speech Data management smart sheets: CONNECTED SPEECH DATA MANAGEMENT SMARTSHEET FOLDER

• All participants follow this structure, independently of the site of treatment (Mexico, Barcelona, Austin, etc.,) (Decision 20241115)

Pre-R01 Connected Speech samples

Before April 2024 all Connected Speech Samples were recorded via ZOOM. These samples are part of the Connected Speech Data Raw but the videos/audios are kept in a separate folder because they were recorded differently, and the format is also different.

The Pre-R01 tasks include:

• STUDY: Therapy trial pre-R01.

  • Picnic Scene 

  • Cat Rescue 

  • Important Event

• STUDY: SpeechFTLD A and B, samples can currently be found here: https://utexas.box.com/s/mu7f0437ls63p22wy4on6f04jxotpgx2  

  • Procedural task Brushing teeth

  • Picnic scene

1. 20251001 Dr.Grasso and Sonia meeting, for Pre-R01 samples we currently have the 1st visit date for each timepoint in the Connected Speech smartsheets for the administration of the Connected Speech samples. We plan on having the exact date of administration for all the timepoints to have more accurate data, but for now we will use those dates that have been extracted from the MADR participant sheet for all the timepoints except for POST, since we don’t have a post timepoint in the MADR participant smartsheet. Jada (RA student) is currently working on adding the post-tx exact dates to the connected speech smartsheets, and we plan on start adding the rest of the dates switching them from the current approximate dates once the post-tx dates are added for both Spanish and Catalan.

Scheme of where the pre-R01 samples are saved:

Open

Overview of Processes/ Procedures

Can the following elements get included in the app? (Y/N/M)

The general overview of processing procedures is as follows:

1. Sample is recorded YES

2. Sample is saved in folder on Box NO- and Maybe not necessary

3. Sample is clipped to ensure no clinician speech or background noise is included Maybe. Probably this could happen OUTSIDE of app and just have an option to re-upload when finished so that the following steps happen automatically. Could diarization take care of this Kesha, has it gotten any better? Then, it could happen in the app and decide to apply diarization so it’s patient-only speech or not!

   1. Cody says in the app there should be a way to clip within app and that way they can stay in app.

      1. Still we want an upload option for audios for older samples

4. a. Audio sample gets run through the following steps

   1. A script that cuts pauses from start and end of samples (Kesha confirmed this is correct and is already in the Acoustic Pipeline) maybe Would definitely like this included. Kesha is this detected via a specific dB level at present? Takes in audio sample and for each audio segment it determines if it’s silent or not. Kesha will look into it

   2. Audio is run through Acoustic Pipeline (Kesha’s script) MAYBE Would definitely like this included

   3. https://www.youtube.com/watch?v=YxZ8cLGWDaE

4. b.Audio sample goes through the following steps

MUCH LATER: Some graphs showing how the person did on each feature relative to controls (or others with the dx) on the same task