Knowledge Base

The Voice Analysis Module

Main Topics

• Introduction to the Voice Analysis Module
• Analyze voice characteristics
• Visualize voice characteristics
• Export voice analysis data

Introduction To The Voice Analysis Module

The Voice Analysis Module enables you to analyze vocal characteristics. FaceReader can currently detect the following emotions in a voice: Neutral, Happy, Sad, Angry. The software was trained and tested using British and American English data, both acted and natural speed. Preliminary tests show potential applicability to other languages, particularly those with close linguistic and cultural similarities.

An audio buffer is used to collect approximately one second of data before starting the analysis because audio (unlike video) cannot be segmented into discrete frames. FaceReader uses a simple Voice Activity Detection (VAD) and analyzes all sound above the threshold. If there is noise in the audio signal this will be analyzed as well.

There is no preprocessing step to decide if audio is speech, so background noise can result in incorrect detections. Using a high-quality microphone and limiting background noise is recommended to improve the accuracy of the voice analysis. Furthermore, it might be necessary to adjust the microphone's sensitivity to achieve the best results.

Please note that you cannot combine the Voice Analysis Module with Baby FaceReader and you cannot analyze emotions in voices in a multi-subject analysis.

Note

The Voice Analysis module employs Voice Activity Detection (VAD) based on a Gaussian Mixture Model (GMM) to distinguish between speech and non-speech segments. This approach helps ensure that only voiced segments are processed, reducing the influence of silence or background noise. While GMM-based VAD is generally reliable under typical conditions, it is not immune to mis-classifications, particularly in environments with overlapping speech, high levels of ambient noise, or non-speech sounds within similar frequency bands. Only if voice is detected, the following measures will be calculated and reported by FaceReader.

Source: FaceReader 10 Reference Manual (Help), Noldus Information Technology