Following the demand of the speech technology market, a number of companies and research laboratories joined their forces in order to produce valuable and reusable resources, especially speech databases. Serving their purpose, the collected databases are used for developing, testing, enhancing and evaluating speech technology products, like interactive voice servers, listening typewriter, speaker verification and identification systems, etc. Especially for capturing intra-speaker variability, the PolyVar database was designed and recorded at IDIAP, as a complement to the Swiss French PolyPhone database, which adresses inter-speaker variability issues. We will detail in the following the specific problems of speech database collection (sampling the speaker population, selection of vocabulary items, ...), and will present actual development we carried out at IDIAP throught the PolyPhone and PolyVar databases.