What statistical modeling tool formed the foundation of the Statistical Parametric Approach emerging in the late 1980s and 1990s?
Hidden Markov Models (HMMs).
The statistical parametric approach marked a significant shift away from physically stitching together recorded units (concatenative synthesis) or explicitly defining every linguistic rule (formant synthesis). This new method utilized statistical models to learn the underlying relationships within speech data. Specifically, researchers employed Hidden Markov Models (HMMs) to statistically model the spectral and prosodic features of speech. Human voice recordings were used to train the HMM, allowing it to learn the statistical distribution linking linguistic input (like duration, pitch, and phonemes) to the resulting acoustic output. This trained model could then generate speech by sampling from the learned distribution, offering flexibility in modifying voice characteristics without needing pre-recorded clips of those specific sounds.
