Auxiliary Loss Function for Target Speech Extraction and Recognition with Weak Supervision Based on Speaker Characteristics.
Katerina ZmolíkováMarc DelcroixDesh RajShinji WatanabeJan CernockýPublished in: Interspeech (2021)
Keyphrases
- loss function
- speech recognition
- automatic speech recognition systems
- recognition engine
- speech sounds
- speaker independent
- speaker dependent
- speaker recognition
- audio visual
- automatic transcription
- speaker verification
- speech recognition systems
- pairwise
- automatic speech recognition
- learning to rank
- support vector
- speech signal
- hidden markov models
- prosodic features
- speaker identification
- risk minimization
- logistic regression
- empirical risk
- hinge loss
- reproducing kernel hilbert space
- regularization term
- speaker diarization
- boosting framework
- stochastic gradient descent
- noisy environments
- vocal tract
- visual speech
- convex loss functions
- feature extraction
- speech synthesis
- acoustic features
- prediction error
- mel frequency cepstral coefficients
- machine learning
- boosting algorithms