Minimum Classification Error Training with Speech Synthesis-Based Regularization for Speech Recognition.
Naoto UmezakiTakumi OkuboHideyuki WatanabeShigeru KatagiriMiho OhsakiPublished in: SPML (2019)
Keyphrases
- speech synthesis
- speech recognition
- minimum classification error
- hidden markov models
- discriminative training
- automatic speech recognition
- statistical pattern recognition
- vocal tract
- speech signal
- isolated word
- pattern recognition
- language model
- text to speech
- classification error
- acoustic models
- speech recognizer
- maximum likelihood estimation
- speech recognition systems
- noisy environments
- gaussian mixture
- training set
- conditional random fields
- training samples
- speaker identification
- speaker independent
- machine learning
- non stationary
- neural network