Abstract: This paper presents the URL-TTS adaptation to the Albayzin Evaluation Campaign within the context of the FALA2010 workshop. The system presented follows the classical scheme of unit selection text-to-speech (US-TTS) systems. URL-TTS presents two distinguishable important features: i) prosody prediction learned from labelled data by means of Case-Based-Reasoning (CBR) and perceptual weight tuning of cost function by means of active interactive Genetic Algorithms (aiGA). Furthermore, the substitution of the classical averaged cost function (AVG) for its root-mean squared (RMS) variant is tested, without finding noticeable improvement on synthetic speech quality. The perceptual tests conducted before submitting the final sentences to the competition obtained acceptable MOS scores (over 3.30).
Index Terms: speech synthesis, unit selection, weight tuning, prosody prediction, interactive genetic algorithms, case-based reasoning.