We present a software tool to assemble a great range of diverse datasets from the Pangloss Collection (a multimedia open archive of under-documented languages). The tool ensures the reproducibility of experiments conducted on these data. As an example, two transcribed audio corpora of Chinese minority languages (Japhug and Na) are proposed, under a Creative Commons license, as reference corpora for experiments in Natural Language Processing, and as examples of a pipeline that can be generalized to other corpora from open archives. An overarching goal of making language archive data available in an easily accessible and usable form is to facilitate the development and deployment of state-of-the-art natural language processing tools for the full range of human languages. This presentation, which follows a previous paper on the same topic, reports on new developments including feedback on a deposit at Hugging Face Datasets. A full-text English version of this paper is available from the following address: https://hal.archives-ouvertes.fr/hal-03846839