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