QuechuaBase/asr-ser-quechua-collao-embeddings

ASR-SER embeddings for Quechua Collao

This repository contains embeddings only. It does not contain raw audio.

These embeddings were extracted for ASR-to-SER transfer experiments on the Quechua Collao emotional speech corpus. The release is intended for reproducible feature-extraction experiments and downstream analysis.

Dataset contents

• One PyTorch tensor per utterance stored as an embedding file under embeddings/
• A sanitized metadata table describing utterance identifiers, speaker codes, emotion labels, and regression targets
• No model checkpoints, training logs, or raw audio are included
• The release contains 12,420 utterance-level embedding files and associated metadata

Encoder and protocol

• Upstream ASR encoder: QuechuaBase/xls-r-cpt-qxp-silver
• Embedding format: PyTorch .pt files with a dictionary containing embedding, utterance_id, speaker_code, emotion_label, arousal, valence, dominance, and duration
• Validation protocol: six-fold leave-one-actor-out evaluation with actor-disjoint train/validation splits
• The embeddings were extracted for ASR-to-SER transfer experiments and are intended for feature extraction and downstream analysis

Source note

The original Quechua Collao emotional speech corpus should be cited according to its original source. The upstream ASR encoder should also be cited separately.

How to load the embeddings

import torch
from pathlib import Path

root = Path("embeddings")
example = torch.load(root / "10001.pt", map_location="cpu")
print(example["utterance_id"])
print(example["embedding"].shape)

Notes

• Raw audio is not included.
• The release is intended for academic and research use.
• Please contact the maintainers before using the data for commercial purposes.

Citation placeholder

If you use this dataset release, please cite both:

1. The original Quechua Collao speech emotion corpus.
2. The upstream ASR encoder used to produce the embeddings.