next up previous contents index
Next: Recommendations on recording strategies Up: Introduction Previous: The communication chain

Specifications for speech corpora production and speech assessment environments


Both approaches are suitable for the generation of spoken language corpora (cf. Chapters 3 to 5): while flawless speech   may be sufficient for allround utilisation, a collection of in-car  speech at certain conditions (car speed, road surface, etc.) may be highly valuable for the example given above, but is restricted to this special case. In both the ideal and the real-life approaches, a detailed protocol of all recording conditions is mandatory (cf. Section 8.7 on reproducibility ).

In designing speech assessment environments, we can find a similar distinction between field  and laboratory  tests (cf. Chapter 12). While a field test  is executed in a real (natural or virtual) environment, laboratory tests  with subjects may utilise both clean and real-life speech. For speech output assessment in case of speech synthesisers  or codecs, the ``clean'' output speech has to be subsequently adopted to real-life conditions by signal processing steps (cf. Section 8.8).


EAGLES SWLG SoftEdition, May 1997. Get the book...