eduzhai > Applied Sciences > Engineering >

Incorporating Broad Phonetic Information for Speech Enhancement

  • king
  • (0) Download
  • 20210506
  • Save

... pages left unread,continue reading

Document pages: 5 pages

Abstract: In noisy conditions, knowing speech contents facilitates listeners to moreeffectively suppress background noise components and to retrieve pure speechsignals. Previous studies have also confirmed the benefits of incorporatingphonetic information in a speech enhancement (SE) system to achieve betterdenoising performance. To obtain the phonetic information, we usually prepare aphoneme-based acoustic model, which is trained using speech waveforms andphoneme labels. Despite performing well in normal noisy conditions, whenoperating in very noisy conditions, however, the recognized phonemes may beerroneous and thus misguide the SE process. To overcome the limitation, thisstudy proposes to incorporate the broad phonetic class (BPC) information intothe SE process. We have investigated three criteria to build the BPC, includingtwo knowledge-based criteria: place and manner of articulatory and onedata-driven criterion. Moreover, the recognition accuracies of BPCs are muchhigher than that of phonemes, thus providing more accurate phonetic informationto guide the SE process under very noisy conditions. Experimental resultsdemonstrate that the proposed SE with the BPC information framework can achievenotable performance improvements over the baseline system and an SE systemusing monophonic information in terms of both speech quality intelligibility onthe TIMIT dataset.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×