Abstract
Since the late 1990's correlates of rhythm classes of languages have been profusely used to search for differences between languages, dialects, speaking styles, degree of foreign accents, etc. Over the years the original attractiveness of the metrics has been replaced with suspicion and, occasionally, even fierce criticism.
Among many reservations the critics argue that the metrics are only based on durational measures ignoring other dimensions of prominence, and they are considerably influenced by local temporal variation in utterances. We argue that the metrics could still be exploited in speech research as long as we do not expect them to reflect "speech rhythm" and as long as the proper account of their use is supplied.
This study provides simulations to demonstrate the behaviour of the most commonly utilised metrics, and presents representative measurements of some Czech and English speech recordings under several conditions.