System and method for training cloned tone and rhythm based on Bottleneck features

摘要: The invention relates to the technical field of voice synthesis, recognition and cloning, provides a cloning implementation scheme based on Bottleneck features (language featuresof audio) by combining synthesis technology, technology transfer learning technology. A training system method are included. TTS service with highnaturalness similarity is provided using small number samples, so that target user characteristics provided, problems large sample size, long manufacturing period high labor cost solved. comprises data acquisition module, an acoustic feature extraction rhythm multi-person module module. further system. steps oftraining corpus preparation, extraction, fine adjustment all modules speech synthesis.