Overleaf is a collaborative cloud-based latex editor. It’s very popular among researchers and students as it allows to collaborate in document writing using Latex, and to compile these documents without having to install packages. However, your documents are in somebody else server, and you have to pay a subscription to collaborate with more than one person. In this article, I’m going to show you how I overcame these problems by hosting my own version of overleaf.

What is KMeans? KMeans is an algorithm that partitions a dataset into K distinct non-overlapping clusters. Every data point is assigned to one and only one cluster. This partitioning is useful for clustering and also for quantization, as we can describe a data point with the index corresponding to the assigned cluster. The algorithm is very simple and aims to minimize the within-cluster variance: Randomly initialize K vectors (centroids) with the same dimensionality as the data.

What is a language model? Language models — like ChatGPT, Llama and Mistral — are a hot topic nowadays and everybody seem to be trying them out; but, what about language models for audio generation? In this article, I’ll show you some of the experiments I did to generate audio using GPT2, EnCodec and EnCodecMAE. But first, let’s talk about language models (LMs). Language models estimate probabilities of sequences of words given a corpus.

Stereo and panning Stereophonic sound (stereo) is a method of sound reproduction where 2 independent channels are used to recreate the multi-directional aspect of sound. If we want a sound to feel placed on the left of the listener, then during the mixing process, we can increase the level of the sound for the left channel and decrease it for the right one. This process of artificially placing sources in the stereo field by assigning different levels to each channel is called panning.