Spoken Keyword Extraction
The application of automatic keyword extraction in the speech modality, aimed at extracting keywords from spoken audio. It faces structural challenges such as disfluencies, repetitions, self-corrections, and prosodic boundary cues. The main approaches are divided into transcription-based methods and end-to-end acoustic methods.
Timeline (last 90 days)
The transcription route for speech KE is limited by the propagation of speech recognition errors, and recognition errors on proper nouns and domain terminology directly harm keyword quality.
The survey authors argue that the transcription route and the end-to-end acoustic route are not substitutes, and that multimodal fusion is the pragmatic direction, while effective multimodal fusion remains to be solved.
All Facts (2)
The transcription route for speech KE is limited by the propagation of speech recognition errors, and recognition errors on proper nouns and domain terminology directly harm keyword quality.
50%UnverifiedThe survey authors argue that the transcription route and the end-to-end acoustic route are not substitutes, and that multimodal fusion is the pragmatic direction, while effective multimodal fusion remains to be solved.
50%