Sounding Out Emotions
For several years now, the automatic analysis of emotions for market research purposes has been a primary research area for the GfK Verein. Working together with engineers at the University of Passau and start-up firm audEERING, the GfK Verein has developed an instrument that detects emotions using the voice.
Voice analysis is now available for GfK customers with Market Builder Voice.
The GfK Verein has already developed sophisticated software in this field. This product, the GfK EMO Scan automatically uses facial expressions to detect emotions. However, visual methods of emotion recognition are not always possible, such as in telephone interviews, for example. Moreover, not all emotional states can be measured through facial expressions. For instance, emotional arousal cannot be inferred from facial expressions. Emotional arousal is an important indicator of the personal relevance of certain experiences and can be easily detected in the voice.
Voice Analytics – “Sounding Out the Consumer’s Voice”
In a mission to expand the range of applications for emotion analysis, the GfK Verein collaborated with engineers at the University of Passau and the start-up firm audEERING to develop an instrument that would recognize emotions through vocal patterns. The developers integrated commercial needs into the product as seamlessly as possible by collaborating with divisional representatives from GfK SE to determine the market requirements of a voice analysis instrument. One of the main decisions during the development process was to have voice recognition focus on recognizing emotional arousal.
The prototype developed for German voice recordings was very good at recognizing emotional arousal. Developers were initially unsure how reliable the voice analytics tool would be in detecting emotions in other languages. Now, the GfK Verein has investigated this question and expanded the software to include voice analysis in German, English, Spanish and Chinese.
Market Builder Voice – A Tool with Practical Application
MarketBuilder Voice is structured around a situation wherein study participants will be presented with new product concepts and then asked to speak freely about their reactions to this product. Their answers will be recorded and evaluated using automatic analysis tools, including the Market Builder Voice. This product stands out because it uses emotional voice analysis to examine “how” something is being said in conjunction with “what” is being said. The two dimensions, content and emotion analysis, are summarized, and together they offer insight into the full degree of emotional influence.
Developers are currently investigating the question of how accurate voice analysis software can be when used in languages for which it has not been programmed or optimized. Once these results have been obtained, the GfK Verein hopes to be able to determine if and to what degree the software needs to be expanded and optimized for other languages, or if the current version is able to adequately measure the emotional content of unfamiliar languages. Even though this research angle has only just begun, initial results suggest that the software will be able to be applied to other languages. For example, algorithms optimized for German, Chinese, and Spanish have already been able to accurately predict emotions in English recordings.
Another important question closely tied to this if whether people perceive the emotions present in a vocal expression independently of whether or not they understand a language? And, if so, how different is their perception of emotions from someone who does understand the language? This will hopefully allow further conclusions to be drawn concerning the capacity for generalizing the software’s analysis results.
Prof. Dr. Björn Schuller
(University of Passau, Imperial College London)