Perception of Social Speaker Characteristics in Synthetic Speech

Rallabandi, Sai Sirisha; Bharadwaj, Abhinav; Naderi, Babak; Möller, Sebastian

doi:10.21437/interspeech.2021-1229

Cited by 1 publication

(3 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The evaluation of these systems have consistently suggested improvements in the existing synthesis procedure [14,15,16,17]. In our previous work, we have studied the commercial TTS systems, Google 1 and Amazon voices 2 [18,19]. Our study shows various speaker attributes contributing to the perception of the universal dimensions (warmth and competence) in synthetic speech [18].…”

Section: Introductionmentioning

confidence: 86%

“…Figure 1 displays the block diagram of the current work. DSSC represents the desired social speaker characteristics from synthetic speech [18]. DAV represents derived acoustic features [19].…”

Section: Overviewmentioning

confidence: 99%

“…In our previous work, we have studied the commercial TTS systems, Google 1 and Amazon voices 2 [18,19]. Our study shows various speaker attributes contributing to the perception of the universal dimensions (warmth and competence) in synthetic speech [18]. In [19], we have derived the acoustic features that could affect warmth and competence in commercial TTS voices using linear regression.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

On incorporating social speaker characteristics in synthetic speech

Rallabandi¹,

Möller²

2022

Preprint

Self Cite

View full text Add to dashboard Cite

In our previous work, we derived the acoustic features, that contribute to the perception of warmth and competence in synthetic speech. As an extension, in our current work, we investigate the impact of the derived vocal features in the generation of the desired characteristics. The acoustic features, spectral flux, F1 mean and F2 mean and their convex combinations were explored for the generation of higher warmth in female speech. The voiced slope, spectral flux, and their convex combinations were investigated for the generation of higher competence in female speech. We have employed a feature quantization approach in the traditional end-to-end tacotron based speech synthesis model. The listening tests have shown that the convex combination of acoustic features displays higher Mean Opinion Scores of warmth and competence when compared to that of individual features.

show abstract