Removing Nuisance Variables from Acoustic Word Embeddings Obtaining - - PowerPoint PPT Presentation

removing nuisance variables from acoustic word embeddings
SMART_READER_LITE
LIVE PREVIEW

Removing Nuisance Variables from Acoustic Word Embeddings Obtaining - - PowerPoint PPT Presentation

Lisa van Staden Removing Nuisance Variables from Acoustic Word Embeddings Obtaining transcriptions is expensive and not always possible. Popular methods for speech processing rely on transcribed speech. 1 Low-Resource Speech and Language


slide-1
SLIDE 1

Removing Nuisance Variables from Acoustic Word Embeddings

Lisa van Staden

slide-2
SLIDE 2

Low-Resource Speech and Language Processing

Popular methods for speech processing rely on transcribed speech. Obtaining transcriptions is expensive and not always possible.

1

slide-3
SLIDE 3

Tasks in LSL Processing

We don’t always need to predict text labels:

  • Query-by-Example Search: search speech using speech.
  • Tasks need speech segments to be compared.

2

slide-4
SLIDE 4

Acoustic Word Embeddings

We want to map speech to these representation without using labels.

3

slide-5
SLIDE 5

Nuisance Variables: Speaker and Sex

Acoustic properties of speech from different speakers/sexes differ. We want embeddings to be robust.

4

slide-6
SLIDE 6

Current Models

5

slide-7
SLIDE 7

What’s Next

  • Improved models: Disentanglement with adverserial training.
  • Using embeddings in downstream tasks.
  • Investigate the phonetic information in embeddings.
  • Links to language acquisition.

6