In this case there are a number of options available.

1. Fitting a linear regression model and looking at the fitted vs residuals plot to ascertain that the residuals are normally distributed and the assumption of homogeneity of variance is not violated. This method succeeds even when the data is not normally distributed as the assumption of normality is for the residuals and not for the data. Here is a good tutorial: https://rpubs.com/iabrady/residual-analysis

2. Failing to fit a linear regression model, you may want to consider a generalised linear regression model with Gamma distribution. This, however, requires that your data is always positive and non-zero. The following tutorial explains this. http://rstudio-pubs-static.s3.amazonaws.com/5691_192685385fc445c9b3fb1619960a20e2.html

3. In most of the cases the above two approaches should work. The variable can be transformed if it can be interpreted on the transformed scale. For example, log transformed noise level may not have an intuitive interpretation.

Hello,

Previously, it was recommended to normalize the data to fit a normal or Gaussian distribution. However, at present several articles have emerged that point out the mistake that is being made when converting the data to a logarithm of 10 or square root, etc. Because the interpretation becomes very confusing.

Please refer this interesting relevant discussion

https://stats.stackexchange.com/questions/6350/anova-assumption-normality-normal-distribution-of-residuals