What Did You Say? A Web-Based Validation of a Speech-In-Noise Task

# What Did You Say? A Web-Based Validation of a Speech-In-Noise Task
## Rutgers University Center for Cognitive Science
### Jason Geller, Ph.D.
### BeOnline 2021

---

---
background-image:url("images/cocktail.jpeg")

---
# Open-Set vs. Closed-set SiN Tasks

+ Open-set

<html>
  <head>
    <title>Audio</title>
  </head>
  <body>
    <script>
      function play() {
        var audio = document.getElementById("audio");
        audio.play();
      }
    </script>
    <input type="button" value="PLAY" onclick="play()">
    <audio id="audio" src="ball_noise.wav"></audio>
  </body>
</html>

---
# Closed vs. Open-Set Tasks
+ Closed-set

- Fall <br>
  - *Ball* <br>
  - Shawl <br>
  - Wall <br>

---
# Issues

- Sentence-based (open-set) tasks are generally preferred as they are the most ecologically valid

- However: 
  + Open-set tasks are difficult to use experimentally
     -  Engages a whole host of processes not related to speech perception
--

- We need a closed-set task that better approximates everyday listening situations

+ Lexical competition
 + Talker variability

---

# Iowa Test of Consonant Perception
+ 4-AFC closed-set (single word) SiN task

- 120 target words
  
      - Each target word belongs to a set and within the set appears as target and foil
      
      - Each word spoken by 4 speakers (2 women)
      
  - Minimal pairs differing by first consonant
  
+ Uses multi-speaker babble

+ All analysis scripts, materials, and data are available at our OSF page.
  - Roll your own ITCP

---

---

---
# Procedure

+ Two sessions (spaced one week apart)

- Gorilla 
  - Prolific

---

# Procedure

+ Session 1 (*N*=199)

---
# Procedure

+ Session 2 (*N* = 98)

---
# Procedure
- Let's try it out!
https://app.gorilla.sc/openmaterials/97811

---
# Pilot

+ Performance in Silence
    - *N* = 50
      - *M* = 95%
---

# Reliability
 + Test-Retest
  - ICC(A,2)  = .80

<img src="BeOnline_JG_files/figure-html/unnamed-chunk-6-1.png" style="display: block; margin: auto;" />
---
# Validity

+ Session 1 of ITCP and CNC

<img src="BeOnline_JG_files/figure-html/unnamed-chunk-7-1.png" style="display: block; margin: auto;" />
---
# Validity
+ Session 1 of ITCP and AzBio

<img src="BeOnline_JG_files/figure-html/unnamed-chunk-8-1.png" style="display: block; margin: auto;" />
---
# IRT Model

---

# Future Directions

- We do have data from 50 participants and data look comparable.
  
+ Online testing for patients (e.g., hearing aid and cochlear implant users)

+ Use it Experimentally
]
--
.pull-right[
<img src="inlab.png" width="4563" style="display: block; margin: auto;" />
]

---

# What Present Me Learned From Past Me

+ Give bonuses for completing second session - *set up separate studies on recruitment platform*

+ Be explicit in your study description

+ Email subjects multiple times to remind them of an upcoming session

+ Try to make experiment length reasonable

---

# Thank You