USM-Lite: Quantization and Sparsity Aware Fine-tuning for Speech Recognition with Universal Speech Models.
Shaojin DingDavid QiuDavid RimYanzhang HeOleg RybakovBo LiRohit PrabhavalkarWeiran WangTara N. SainathShivani AgrawalZhonglin HanJian LiAmir YazdanbakhshPublished in: CoRR (2023)
Keyphrases
- speech recognition
- fine tuning
- acoustic models
- speech signal
- speech synthesis
- hidden markov models
- language model
- speech recognizer
- automatic speech recognition
- speaker identification
- speech processing
- pattern recognition
- speech recognition technology
- recognition engine
- speaker dependent
- probabilistic model
- word error rate
- speech recognition systems
- speech recognizers
- cepstral coefficients
- keyword spotting
- noisy speech
- computer vision
- noisy environments
- speaker recognition
- broadcast news
- feature space
- information retrieval