Week 10 – Final Steps

Overview 

This week, I set about implementing the final changes to my models and worked toward my final product. I also tested the model with a variety of different videos with different colors to evaluate the model’s performance.   

Audio Alteration 

Last week, I mentioned I set about a different method for differentiating between different sounds and actual static. That method in particular was recording how many times a given sound signal spiked above a threshold frequency.  

As you can see from the sound graph above, there are various segments of rapid up and down movement, almost making an orange block. These periods represent sounds that the model flags as noise. The shorter segments represent voices that the model has no problem recognizing as separate sounds. My goal is to have the model recognize the orange blocks as actual sounds as well and will do so by relating them with the number of times the signal spikes above a certain threshold. In my last week, I seek to finally find a solution to this problem.  

Colorization 

Last week, I also mentioned that my model is able to work with images of a dimension of 512×512 pixels. Videos produced from this resolution are much more clear and have much better colorization than previous videos which were created from images of 256×256 pixels. Since most old videos are produced with low resolution, to begin with, there is not much need to have my model run on higher and higher resolution. This means my colorization model is officially complete.   

Next Steps 

With colorization completed, my main objective for this final week is audio alteration. I will attempt to code the differentiation between sounds and static with the method I described in this post. Hopefully, the issue will be resolved by the time of the presentations next week!