On-Device Speaker Anonymization of Acoustic Embeddings for ASR based on Flexible Location Gradient Reversal Layer.
Md Asif JalalPablo Peso ParadaJisi ZhangMete OzayKarthikeyan SaravananMyoungji HanJungin LeeSeokyeong JungPublished in: INTERSPEECH (2023)
Keyphrases
- automatic speech recognition
- speech recognition
- acoustic features
- speech sounds
- acoustic models
- speech recognizers
- speaker independent
- speech signal
- prosodic features
- speech segments
- privacy preserving
- speech recognition systems
- hidden markov models
- spontaneous speech
- multi layer
- information loss
- location information
- privacy preservation
- edge detection
- neural network
- privacy protection
- lightweight
- mel frequency cepstral coefficients
- vocal tract
- speech recognizer
- source localization
- audio visual
- speaker recognition
- speaker verification
- manifold learning
- data quality
- vector space
- privacy preserving data publishing
- speaker identification
- broadcast news
- noisy environments