Semi-supervised acoustic modelling for five-lingual code-switched ASR using automatically-segmented soap opera speech.
Nick WilkinsonAstik BiswasEmre YilmazFebe de WetEwald van der WesthuizenThomas R. NieslerPublished in: CoRR (2020)
Keyphrases
- automatic speech recognition
- automatically segmented
- semi supervised
- speech sounds
- speech recognition
- acoustic features
- speech signal
- speech recognition systems
- emotion recognition
- speech recognizers
- acoustic models
- emotional speech
- speech segments
- speaker independent
- word error rate
- semi supervised learning
- web services
- broadcast news
- speech synthesis
- noisy environments
- spontaneous speech
- formant frequencies
- speech recognizer
- speech corpus
- spoken words
- unlabeled data
- hidden markov models
- conversational speech
- prosodic features
- source code
- description language
- labeled data
- active learning
- recognition errors
- supervised learning
- pairwise
- semi supervised clustering
- vocal tract
- probabilistic model
- language model
- multi view
- speaker recognition
- sound source
- speaker identification
- text to speech
- speaker verification
- skin lesion
- audio visual
- back end
- training data