Quechua speech emotion recognition datasets

KuralHub lists 1 speech emotion recognition dataset for Quechua. Each entry below gives the emotion labels, size, recording setup, licence, where to get it and how to cite it. Quechua is also part of the benchmark. The best of 11 models reached 83.0% test accuracy (whisper-large).

Language family: Quechuan
ISO code: qu
Spoken mainly in: Throughout the central Andes Mountains including Argentina, Bolivia, Chile, Colombia, Ecuador, and Peru
Datasets listed: 1
In benchmark: Yes

Datasets at a glance

Dataset	Type	Samples	Speakers	Recording	Access	Licence
Quechua Collao Corpus	Audio	12420	6	Acted	Open	CC BY 4.0

Benchmark results

Quechua is one of the 29 languages in the KuralHub benchmark. Each of the 11 frozen speech models was trained with a small classification head on the shared emotions and scored on a held-out test split. Over all 11 runs the mean test accuracy was 58.7% (median 71.0%). The best run reached 83.0% with whisper-large. Compare with other languages.

**Quechua Collao Corpus** (run ID `Quechua-Collao-Corpus`). Scores on the test split, best run highlighted.
Model	Family	Val acc.	Test acc.	Macro-F1	Weighted-F1	Logs
hubert-base-ls960	HuBERT	0.81	0.80	0.79	0.80	logs
hubert-large-ls960-ft	HuBERT	0.50	0.50	0.47	0.47	logs
wav2vec2-base	wav2vec 2.0	0.78	0.76	0.76	0.76	logs
wav2vec2-large-960h	wav2vec 2.0	0.39	0.36	0.33	0.33	logs
wav2vec2-large-lv60	wav2vec 2.0	0.22	0.22	0.18	0.18	logs
wav2vec2-xls-r-300m	wav2vec 2.0	0.43	0.43	0.43	0.43	logs
wav2vec2-xls-r-1b	wav2vec 2.0	0.28	0.30	0.23	0.23	logs
wavlm-base-plus	WavLM	0.75	0.71	0.70	0.71	logs
wavlm-large	WavLM	0.76	0.73	0.73	0.73	logs
whisper-small	Whisper	0.82	0.82	0.81	0.82	logs
whisper-largebest	Whisper	0.84	0.83	0.83	0.83	logs

Quechua Collao Corpus

A speech corpus of Quechua Collao for automatic dimensional emotion recognition

Type: Audio
Emotion labels: Happy, Sad, Bored, Fear, Sleepy, Calm, Excited, Angry, and Neutral
Size: Number of samples: 12420, speakers: 6, duration: 15 hours
Speakers: 6
Recording: Acted
Access: Open
Licence: CC BY 4.0
File format: .wav
Website: Dataset Homepage
Paper: Paper Link
Download: Download
Contact: rpaccotacya@unsa.edu.pe

The dataset comprises recordings of words and sentences in the Quechua Collao language. These utterances were specifically collected to represent nine categorical emotions, aiming to facilitate research in automatic emotion recognition systems for this language. Quechua Collao corpus for automatic emotion recognition in speech. Audios are provided, alongside csv files with labels from 4 annotators for valence, arousal, and dominance values, using a 1 to 5 scale. Categorical labels are also included, as well as the script used for recording. This script contains sets of words and sentences written in Quechua Collao for 9 emotions.

Used in the KuralHub benchmark (run ID Quechua-Collao-Corpus). See the results above.

Citation

@article{paccotacya2022speech,
  title={A speech corpus of quechua collao for automatic dimensional emotion recognition},
  author={Paccotacya-Yanque, Rosa YG and Huanca-Anquise, Candy A and Escalante-Calcina, Judith and Ramos-Lov{\'o}n, Wilber R and Cuno-Parari, {\'A}lvaro E},
  journal={Scientific Data},
  volume={9},
  number={1},
  pages={778},
  year={2022},
  publisher={Nature Publishing Group UK London}
}

Source: datasets/Quechua/Quechua-Collao.md · Suggest an edit

Know another Quechua speech emotion dataset, or a newer version of one listed here? Suggest it or open a pull request.

This page is generated from datasets/Quechua/ in the KuralHub repository. If you use a dataset, cite its authors. If KuralHub helped you find it, please also cite the KuralHub paper.