In Angami, Ao, and Mizo, pre-trained wav2vec2 models recognize tones best from middle layers, with Mizo highest (88% F1) and Angami lowest (61% F1).
Layer-wise analysis of a self-supervised speech representation model
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
background 1
citation-polarity summary
fields
eess.AS 1years
2025 1verdicts
CONDITIONAL 1roles
background 1polarities
support 1representative citing papers
citing papers explorer
-
Tone recognition in low-resource languages of North-East India: peeling the layers of SSL-based speech models
In Angami, Ao, and Mizo, pre-trained wav2vec2 models recognize tones best from middle layers, with Mizo highest (88% F1) and Angami lowest (61% F1).