Survey 08/15/2021
We surveyed 204 random adult consumers from the U.S during August 11-12, 2021. Survey respondents were recruited from a large national survey panel. The survey availability was posted to the panel, and interested respondents were able to take up the survey. Panel members were told that the survey would involve listening to a voice recording and then answering questions about it. The survey was conducted as a blind split test. Respondents were randomly assigned to listen to one of two voice recordings. They were then asked to score various qualities of the voice they heard and how appealing the voice would be in a number of different applications. Possible scores were 0 to 10. Voice A: Normal TTS from Microsoft AZUR Voice B: The same TTS controlled with our Emotion Synthesis Tech – EMS-AZUR
Voice A: Normal TTS from Microsoft AZUR:
Voice B: The same TTS controlled with our Emotion Synthesis Tech – EMS-AZUR:
Row Data Set Results:
After Cleaning The Data Set:
Methodology: Individual responses that were outliers or self-contradictory were excluded from the data – e.g. respondents who scored everything as 10’s or 0’s or who scored the voice as both highly pleasant and highly annoying. These types of responses indicate that the respondent was not giving careful consideration to the task at hand.
Results: 32% Improvement in caregiver environment and 22% more satisfying in general when Emoshape controls Microsoft Neural TTS in real-time.
