XLS-R fine-tuning on noisy word boundaries for unsupervised speech segmentation into words.
Robin AlgayresPablo Diego-SimonBenoît SagotEmmanuel DupouxPublished in: CoRR (2023)
Keyphrases
- fine tuning
- word segmentation
- word recognition
- cursive script
- numeral strings
- chinese word segmentation
- noisy environments
- syntactic categories
- speech recognition systems
- pos tagging
- n gram
- english words
- out of vocabulary
- speech recognition
- related words
- spoken document retrieval
- viable alternative
- automatic transcription
- fine tune
- speech segments
- lexical features
- segmentation algorithm
- spontaneous speech
- word sense disambiguation
- recognition errors
- unknown words
- pointwise mutual information
- handwriting recognition
- automatic speech recognition
- word meaning
- grapheme to phoneme conversion
- word level
- fully unsupervised
- fine tuned
- word frequencies
- word pairs
- conversational speech
- segmentation method
- word error rate
- english text
- spoken documents
- image segmentation
- text corpus
- multiword
- speech signal
- stop words
- lexical semantics
- speech recognizer
- level set
- broadcast news
- linguistic information
- linguistic knowledge
- speech corpus
- lexical information
- speech sounds
- language independent
- word sense
- text lines
- text input
- character recognition
- text documents
- wordnet
- text classification
- text to speech
- co occurrence