masked_prosody_model

A transformer-based model for prosody prediction with masked inputs. This model processes pitch, energy, and voice activity detection features to predict prosodic patterns in speech.

Installation

pip install masked_prosody_model

Note: torch and torchaudio need to be installed separately.

Usage

from masked_prosody_model import MaskedProsodyModel

# Load the pretrained model
model = MaskedProsodyModel.from_pretrained("cdminix/masked_prosody_model")

# Process an audio file and get representations
representation = model.process_audio("some_audio.wav", layer=7)  # layer between 0 and 15, 7 was used in the paper

Acknowledgments

This model was trained using Cloud TPUs supplied by Google's TPU Research Cloud (TRC). We thank them for their support.