This archive contains speech samples generated by team IRISA/EXPRESSION corpus based TTS system, in the beginning of march 2014. 

Each folder contains 2 sentences, randomly selected within a set of 100, in order to show the speech quality obtained using different parameters for the speech synthesis system. 

The sentences are the n°23 and 25 from our test corpus named Combescure. This corpus contains sentences from a different speech register than the one used for the TTS corpus recordings and possess the property of being phonetically balanced. The content of each sentence, in French language is: 

N°23: "Bien sûr, je connais son nom." (Of course, I know his name.)
N°25: "Désormais, je me tournerai quant-il partira." (Henceforth, I will turn away when he will leave.)

Speech samples in folders 1cov2pho, 1cov2phoComp, 1covAll and astar were generated using a A* algorithm for unit selection process. Samples in folders viterbi10, viterbi100 and viterbi1000 were generated using a Beam-search Viterbi algorithm with a beam of size 10, 100 and 1000, respectively. 

Folder 1cov2pho contains sentences generated using a TTS corpus featuring only a 2-grams covering. 
Sentences in 1cov2phoComp were generated using a TTS corpus obtained with 2-gram covering completed with random sentences.
The corpus used for 1covAll was created with a covering of the features used by the unit selection algorithm. 

The remaining folders - astar, viterbi10, viterbi100 and viterbi1000 - contain speech generated with the full TTS corpus (7h).

