I build a system once for a furby-like product that was designed so that the voice talent could listen to all sorts of speech snippets, and, crucially, play the

I build a system once for a furby-like product that was designed so that the voice talent could listen to all sorts of speech snippets, and, crucially, play them back over the exact same hardware as the final toy's. These 'golden ear' folks would pick the best (heavily compressed) speech that would be used in-product.