Heliumspeech Unscrambling Method Based on Spectrogram Lexicon Learning

In order to unscramble heliumspeech effectively, a novel heliumspeech unscrambling method is proposed, which consists of correction network and unscrambling network. First, a common working language lexicon is established and used to generate supervision signals and vector signals of the correction network. The correction network learns the spectrogram features of heliumspeech as well as normal speech from the supervision signals and vector signals to obtain the correction network parameters. Then, the correction network parameter set with the largest spectrogram fitness is used to correct heliumspeech, according to which, the vector signals and supervision signals are generated. Finally, the unscrambling network unscramble the corrected heliumspeech. Therefore, the individual spectrogram features of the heliumspeech signal of divers in different environments are fully used to improve the learning efficiency of the machine learning networks. Simulation results show that the method proposed is effective.