Speech and speech releated resources at BAS
Florian Schiel · 1998
The Bavarian Archive for Speech Signals (BAS) located at the Ludwig Maximilians Universit at M unchen, Germany collects, evaluates, produces and disseminates German speech resources to the scientic community.Our focus is the German language covering a large geographical part of central Europe.Speech and speech-related resources are usually produced for certain tasks or projects.Therefore, it is not easy for scientists or engineers starting a new project or application to decide, whether existing resources may be re-used for their special purpose, or whether it is necessary to nance an new specialised data collection (which is usually very expensive).With this contribution we'll try to facilitate this decision by giving detailed information about existing resources as well as possibilities to produce new resources.This paper has two major parts.The rst part deals with our experiences during the last three years to produce new, highly re-usable speech resources in close cooperation with industrial partners.We will explain our BAS policies that ensures that valuable resources do not only satisfy shortterm needs of certain industrial applications but are useful for other research or engineering activities as well.Some example projects will high-light these points.Finally we encourage other speech c e n tres in Europe (for other languages than German) to adopt these policies.In the second part we will give a concise description of all existing German speech and speech-related resources that are currently available at BAS.The description will not only give details about the properties of the corpus, a quality assessment in regard to possible applications, but also include an outline under what circumstances the original resource was produced.The latter information might b e v ery important to answer special questions that are usually not contained in the standard description of a speech resource.