For those participants who have tried assistive equalisers before, we saw a strong consensus among the respondents for wanting to understand how the model derives its recommendations. Looking at the quantitative data for the variable Explanability, 79% of participants said knowing how the AI derives the recommended parameter settings is important. The desire for understandable and transparent AI recommendations was not affected by the participants' age, skill type, or level of expertise. The qualitative data confirm these findings, as multiple users did not like the lack of transparency and explainability in the EQ they had tried, as suggested by the user comments.  

• "Would love some explanations of what is corrected by AI and why."

• "There is no particular informative reasoning provided for the recommendation. 'Black Box' of secret sauce."

• "The "why" this setting was chosen. An AI explaining the decision made would be useful. Even if I can guess it from the settings".

 

Based on the quantitative data on the Net Promoter Score metric, the majority of the users would happily recommend the assistive equalisers to a friend or colleague. This result suggests that participants have a positive attitude towards this technology and thought it could add value to their existing workflow. This interpretation is further supported by the fact that 62% of the participants agreed that using the assistive EQ increased their productivity, and 55% found using the Assistive EQ useful. Below are some comments derived from the qualitative data that confirm the reasons participants responded positively to these questions:

• "Speeding up workflow and sometimes correcting things I might not."

• "The speed to use as a starting point when equalising"

• "I like the velocity of workflow."

• "The speed to use as a starting point when equalising"

 

The qualitative data also give us insights into why participants responded neutrally or negatively to some questions. Our interpretation and findings show that there are three critical reasons for the neutral and negative responses that are supported by the user comments below:

Topic 1: the AI does not adapt to the users' style

• "Personal taste is very important, auto eq doesn't work in this way."

• "Too much "mystic" and often not adjusted to my own music style"

• "it's impossible to get the assistive equaliser to give you the exact sound you are looking for."

 

Topic 2: it makes bad suggestions at times

• "It doesn't respect the material by going too far in its settings."

• "I have got some good results on single instruments, but when I applied it to a mix, the results were disappointing."

• "Frequency bands and amount of gain applied are sometimes completely missing."

​

and 

 

Topic 3: it can be time-consuming to rectify the errors 

• It's quicker and easier to set things manually than it is to wrestle the AI.

• Sometimes it is impossible to tweak the recommended settings, so I resort to other EQ plug-ins.

​

The above response might also help explain the neutral and negative responses we received about the other three UX metrics, feeling in control, Smart EQ performance, and making participants feel more confident in applying EQ.

​

Given that users, when processing sound using equalisation make subjective and creative decisions that take multiple variables into account, such as stylistic (e.g., music genre) and the musical context (i.e., how the sound fits in the overall mix) and the AI equalisers cannot take these aspects into account yet, giving multiple eq settings suggestions to the user to choose from and giving complete control to customise AI suggestions and rectify errors is very import. Additionally, applying the suggestions to the regular traditional EQ, with a familiar user interface (e.g., EQ graph), is essential for maintaining consistency with existing users' mental models making the technology easier to use and increasing its usefulness. Violating these factors could negatively impact user acceptance, usefulness, and the adoption and success of the product. The qualitative data show that many positive comments were related to the controllability of the Assistive EQ, for instance.

​

User comments suggesting the importance and appreciation of controllability:

• Since most are suggestive, I can tweak the settings to taste. It's quite useful, indeed.

• I can tweak the settings to taste.

• The ability to have a valid starting point for my manipulation of the sound.

• I like picking the right section of music where it scans and bases the settings from.

• Being able to reference other mixes and suggest settings for my own mix.

• They were perfect 9/10 times, just requiring a little tweaking in some cases.

​

The hypothesis tests we ran on the participants' quantitative data for the UX metric feeling more confident; we saw that amateur users are way more likely to feel more competent and confident using an AI-assisted equaliser than advanced users. Regarding the UX metric Trust, measured by the Likert statement, "I trust the suggestions of assistive equalisers which I have tried",  we found that musicians are significantly more likely to use the EQ suggestion than engineers. Moreover, the data suggest that creative users are significantly more likely to be influenced by AI suggestions than engineers. Finally, amateur users would be more likely to recommend AI-assisted EQ to a friend or colleague than advanced users. These findings do not surprise me and, in a way, confirm our pre-existing assumptions. Essentially, the group differences we were able to measure and reveal through the statistical tests are because engineers and advanced users are more likely to be competent and skilful in using EQs than musicians, creatives, and amateur users. Hence, the engineers and advanced users' competency makes them less likely to trust or accept the parameter settings of the AI. Consequently, these user groups are less likely to recommend such products to friends or colleagues.

​

For those participants who have not tried an assistive equaliser, we saw a strong consensus among the respondents for wanting to understand how the model derives its recommendation. Looking at the quantitative data for the variable Explanability, 79% of participants said knowing how the AI derives the recommended parameter settings is important to them. The ANOVA analysis model showed significant differences regarding the effect explainability as a result of the participants' skill type, expertise, and age. Posthoc tests revealed that beginners and musicians care less about explainability than professional and engineer consumers. The majority of the consumers ( i.e. 71%) who had not tried an AI-assisted equaliser are positive about using such a tool. The most prominent anticipated benefits of using such tools are improved productivity and increased confidence levels that the equalisation setting they applied to process the sound is appropriate. These benefits are shared by all participants irrespective of their skill level or type of expertise. A large percentage (46%) reports that they would most likely trust the AI recommendation.