Life Tech Digital health, AI

New software to help measure the severity of speech impairment

Fill out the form to download the document

For people suffering from speech disorders (e.g. ENT cancers), measuring the severity of speech impairment makes it possible to quantify the seriousness of the disorders and also to monitor the impact of treatment, whether positive or negative, on speech.
 

Competitive advantages

  • Automatic & fast: automatic production of the speech intelligibility score
  • Objective: analysis does not become accustomed to the patient and their particularities
  • Reproducible & easy-to-obtain analyses using a consumer mobile device such as a tablet
  • Follow-up of intelligibility over time for correlation with treatments, operations, rehabilitation, etc.
     

Application

  • Speech disorders
     

Intellectual property

  • Software

Development stage

Demonstration of the technology in a real environment
 

Laboratories

  • IRIT UT3
  • Toulouse University Hospital

The technology is a digital tool to help measure the severity of speech impairment in ENT cancer patients by automatically assessing speech intelligibility.
Diagnosis and monitoring are carried out on the tablet by healthcare staff and the patient.

Measurement of the severity of the disorder usually measured via speech intelligibility based on:

  • Ease of understanding and interpreting words spoken by a person
  • Factors such as clarity of voice, speed of speech, volume, articulation, etc.

Current assessment:

  • Test consisting of the patient reading a reference text and the clinician giving a score based on their hearing perception on a scale, providing an intelligibility score (0 to 10)
    • By a clinician listening to the patient => Loss of objectivity in the analysis
    • By a consortium of experts => High cost and difficult to implement
       

Technical specifications

  • Software tested in real-life conditions with 25 patients with speech disorders
  • Accuracy equivalent to an assessment by a consortium of experts
  • Average error of 1.21 (equivalent to an average difference in score between different experts on the same sample)
  • Over 83% correlation between automatically predicted scores and scores provided by experts on a sample acquired in a real-life situation