The robustness of this task can be improved by applying transforms like SPLICE and CMVN. An architecture of composed weighted finite-state transducers is used for this task by the open-source package Kaldi, which does not support the frequently-used CTC loss function in alignment during this task. Classical algorithms for this task extract 39 MFCC features (-5[2])for each time frame of a sliding window, then feed them (-5[1])into (-5[1])an (*) GMM-HMM. (-5[1])Context-dependent models are used in this task to account for allophones. (10[2]-5[2])Probabilistic methods (10[1]-5[1])for this task (10[1])compute (-5[1])the product of an acoustic (10[1])model (-5[1])and a language model (10[1])to select (-5[1])the most likely word sequence (10[1]-5[1])given a sound signal. (10[1]-5[3])For 10 points, speech synthesis (10[2])is the reverse (10[1])of what natural language processing task used by digital assistants (10[1])like Siri (-5[1])to process user input? (10[3])■END■ (10[9]0[7])