Choosing Your Product's Voice
In this article, I explain my psychology dissertation research on voice stereotypy. I then present my key takeaways for choosing voices for a product based on this research.
Voice Stereotypy refers to the phenomenon of attributing physical and personality traits to people solely based on their voices. When I came across this phenomenon, the implications this could have on the design of voice interfaces intrigued the designer in me. As a result, when I was choosing a research area for my master's dissertation, this became my topic of choice. Now that I had finalized my area of research, the next step was to figure out which dimension of voice stereotypy I wanted to explore. With this goal in mind, I started reading about the studies that were already attempted on this topic.
​
The pioneering researchers on this topic had presented speech samples to participants and asked them to rate the voices on certain personality traits. The findings from their studies indicated that speech contains cues through which listeners can make identical judgements about a speaker's leadership abilities, extroversion levels, kindness etc. However, it was not clear if these judgements were based on the voice or if they were based on the way the speaker delivered their sentences. This was clarified later by a 2018 study which demonstrated that the ratings given by participants after listening to long speech samples were identical to the ratings given after listening to short ambiguous words. This study proved that voices do contain cues through which listeners can make identical judgements about a speaker's personality traits such as warmth, trustworthiness, confidence etc.
​
There were also some studies that successfully mapped the acoustic parameters of a voice with these stereotypical judgements. For instance, a 2007 study identified that male voices with a lower fundamental frequency and formant dispersion were perceived to be more dominant. Similarly, a 2015 study found that voices that had a lower pitch were perceived to be competent and stronger. However, a study that interested me the most was a study from 1981 by Cox & Cooper. In their study, they attempted to explore the real-world implications of voice stereotypy by asking participants to pick a voice that they think is the most suitable for making telephonic announcements. Even though they were not able to arrive at any concrete findings, their work made me gravitate towards wanting to explore the effect of voice-stereotypy in a real-world scenario.
​
I discussed this idea with my supervisor and got their approval for taking this forward. Since my research participants were mostly going to be students from my university, we decided to study the effect of voice stereotypy on how students pick their teachers for voice-based learning. However, instead of taking an open-ended approach towards this investigation, we decided to use an existing stereotype that students have when choosing teachers and check if it is applicable for voice-based learning also.
As my supervisor was already researching gender biases and stereotypes, we decided to focus on gender-related biases that students have about teachers. A study from 2018 demonstrated that students show a bias against female teachers who teach subjects that have a high mathematical content. This seemed to be stemming from an irrational stereotype that women are not as good as men in their mathematical abilities. We decided to explore if this stereotype affects how students choose teachers to learn subjects with high mathematical content on a voice-based medium.
​
We set up an online experiment where 117 randomly recruited participants were presented with brief overviews of 8 academic topics. 4 topics had mathematical content and 4 topics did not have mathematical content. The participants were given an option to choose between a male voice and a female voice to learn more about each topic. The participants repeated this study for 3 pairs of male and female voices in a random order to make sure the results were not an outcome of chance. In order to make sure the voices were balanced, they were paired so that they were equal in terms of dominance, trustworthiness and warmth based on the ratings from an earlier study. The consent of the participants for anonymous participation was obtained before the study. However, they were informed about the intent of this study only after they completed the survey.
​
Our findings indicated that irrational gender-based stereotypes that students possessed about teachers had an effect on the way they choose a voice-based tutor as well. The number of times each participant chose a male voice and a female voice for subjects with and without mathematical content in their 24 trials were converted into percentage values for calculating their mean preference. The results of the study were as follows. The participants showed a stronger preference to learn more about all 8 topics from male voices (59.8% Vs. 40.2%). However, this preference was stronger for topics with math content (63.6% Vs 36.4%) as opposed to topics with no math content (56.07% Vs 43.93). The summary of the results can be seen in the graph below. However, it should be noted that the results of my study need to be replicated with a larger sample size before being published.
My Takeaways
After submitting my dissertation, I began analyzing if my research had any implications on the design of voice interfaces. I reviewed my results along with all the literature I came across through the lens of a designer. My review gave me the following takeaways.
1. Choose a Voice That Reflects the Personality of the Interface
Studies have demonstrated that acoustic parameters such as pitch and frequency of a voice can be linked to the personality traits expressed by the voice. This implies that when choosing a voice for an interface, it could be beneficial to synthesize a voice that expresses the personality traits that the users expect from the interface. For instance, a personal assistant interface could benefit from having a diligent personality whereas a meditation trainer interface could benefit from having a calm personality.
2. Pay Attention to Modulations
When a voice interface interacts with its user, the modulation used for communicating messages will play a vital role in how the user receives the message. A 2017 study asked participants to rate various utterances of "hello" on a scale of trustworthiness. The results indicated that the hellos that were considered to be highly trustworthy were characterized by a high starting fundamental frequency followed by a decrease in frequency at mid-utterance followed up with a high rise in frequency to the finish.
This implies that when designing voice interfaces, attention should be given to how different messages are modulated by the interface. Sounding cheerful when saying 'hello' and sounding apologetic when saying 'sorry' could contribute greatly to the user's experience of using the voice interface.
3. Offer Diverse Voice Options
As mentioned earlier, it is important to choose a voice that reflects the personality of the interface. At the same time, it is also important to give the user the ability to choose from a diverse set of voices that express the required personality. Even though my study indicated that irrational stereotypes might lead people to choose one voice over the other, there were still a significant number of participants (36.4%) whos choices were not inline with the stereotype. This stands testimony to the fact that designers should not use stereotypes to make choices for their users. As designers, promoting diversity to accommodate the choices of all types of users should be our priority. Voices that represent various genders, accents etc. should be offered wherever applicable.
