# Speech error detection with degraded auditory representations

> **NIH NIH F32** · UNIVERSITY OF WISCONSIN-MADISON · 2020 · $37,513

## Abstract

Project Summary/Abstract
 Speakers correct their speech and stay on-target while talking by detecting even small deviations in their
auditory feedback. Access to the acoustic details in auditory feedback crucially depends on intact hearing and
self-perception systems, which can be degraded by lack of language experience, hearing impairment, or, in the
case of cochlear implants, by the implant itself. These degraded auditory representations may compromise the
speech production system, contributing to the loss of intelligibility seen in these populations.
 The suite of experiments proposed here tests how speakers detect deviations in their own speech under two
atypical speaking conditions: speaking a foreign language and speaking with cochlear implants. In second lan-
guage learning, a lack of experience with the foreign language results in an underdeveloped second-language
perception system, which interferes with production of native-like speech. Without a strong acoustic representa-
tion of the speech target for comparison, the perception system is not able to evaluate speech output for errors.
In this experiment, established neuroimaging (magnetoencephalography) and behavioral methods will be used
to assess how people speaking French as a second language use their auditory feedback to assess their pro-
ductions. In people speaking their native language, deviation from an acoustic target is correlated with changes
in acoustic trajectories as well as activity in auditory cortex. These measures will be compared with an offline
self-perceptual measure of acoustic sensitivity to determine how speakers' error-detection differs between a first
and second language.
 In people with cochlear implants, the processor collapses spectral distinctions in the speech input, so that
these speakers do not have access to all of the rich acoustic information available in the speech signal. In parallel
with second language learners, this signal degradation may prevent comparison between the speech output and
speech target, possibly diminishing the ability to correct nascent errors. In this experiment, recorded productions
of CI users' speech will be altered and presented back to them in a discrimination task. The experiment tests
how CI users perceive their own speech and determines whether the degraded speech signal provides enough
information to allow for error-detection in running speech.
 The experiments proposed here seek to understand how auditory feedback affects error-detection when part
of the speech production-feedback loop receives a degraded signal, and the methods will provide the candidate
with the training required to apply linguistic theories of speech production to biomedicine. A better understanding
of how error-detection changes when access to high-fidelity auditory feedback is diminished will provide new
avenues for improved treatments in disorders that affect speech communication.

## Key facts

- **NIH application ID:** 9933773
- **Project number:** 5F32DC017653-02
- **Recipient organization:** UNIVERSITY OF WISCONSIN-MADISON
- **Principal Investigator:** Sarah Bakst
- **Activity code:** F32 (R01, R21, SBIR, etc.)
- **Funding institute:** NIH
- **Fiscal year:** 2020
- **Award amount:** $37,513
- **Award type:** 5
- **Project period:** 2019-05-01 → 2020-10-01

## Primary source

NIH RePORTER: https://reporter.nih.gov/project-details/9933773

## Citation

> US National Institutes of Health, RePORTER application 9933773, Speech error detection with degraded auditory representations (5F32DC017653-02). Retrieved via AI Analytics 2026-05-23 from https://api.ai-analytics.org/grant/nih/9933773. Licensed CC0.

---

*[NIH grants dataset](/datasets/nih-grants) · CC0 1.0*