You mean with 1-D signals? Autoencoders can be used with signals or many signals. Example is denoising motion capture data, where you have locations of different markers attached to person, and you want to reduce noise.
Basically, an autoencoder is a neural network with three components, viz., the encoder, the code, and the decoder. The primary objective of an autoencoder is to capture the underlying data codings in an unsupervised manner.
In an oversimplified form, you may consider it as a feed forward network where the output layer is replaced by the input layer itself.
So, as far as the type of the dataset is concerned, it works well with any kind of data such as gene expression, text, audio or any other.
However, if you are considering the application of Convolutional autoencoder, then its applicability is same that of ConvNets.
Csv aka. Comma separated value is a file format to save the data, not a data form itself. If you don't have 2-dimensional signals as that of image, use linear layers.
In fact, you may encounter some cases where, images are also stored in csv. If you worry about how the values from csv would be fed directly into the network, then you have to convert them in a proper format (in particular case for tensorflow, its array).