Speech emotion recognition benchmark: HuBERT, wav2vec 2.0, WavLM and Whisper in 29 languages

Test scores for 11 frozen speech models on 33 speech emotion recognition datasets in 29 languages: 363 runs in all.

Every run uses the same recipe: the pretrained backbone stays frozen, its last-layer features are averaged over time, and a two-layer classification head is trained on a 70/15/15 split of the dataset, using the five shared emotions (anger, happiness, sadness, neutral and fear) that the dataset contains. See section 3 of the paper for details and section 4 for the analysis.

Accuracy: share of test clips classified correctly.
Macro-F1: F1 averaged over emotion classes, each class counted equally.
Weighted-F1: F1 averaged over classes, weighted by how often each class appears.

Download: runs.csv · runs.json · per-run logs on GitHub.

Want another model in these tables? Suggest a checkpoint, or run it with the same recipe and send the logs (how to). New datasets for the benchmark are welcome too.

Best model per language

The single best run for each language. Table 2 in the paper names the best model family per language but does not say how that table was aggregated. For 7 of the 29 languages (English, French, Hindi, Japanese, Kannada, Odia and Urdu) the family of the single best run differs from Table 2.

**Best single run per language.** Test accuracy (%) and macro-F1 of the top run, next to the mean over all of that language's runs.
Language	Family	Best model	Dataset	Test acc.	Macro-F1	Mean acc.
Afrikaans	Indo-European	whisper-small	AfrikaansSE	71.0	0.70	52.4
Amharic	Afro-Asiatic	whisper-small	Amharic Speech Emotion Dataset (ASED)	90.0	0.90	63.6
Arabic	Afro-Asiatic	whisper-large	ANAD	82.0	0.81	66.1
Bengali	Indo-European	whisper-large	SUBESCO	89.0	0.89	46.0
Chinese	Sino-Tibetan	whisper-large	ESD	98.0	0.98	75.5
English	Indo-European	hubert-base-ls960	IESC	82.0	0.81	51.0
French	Indo-European	wavlm-large	CaFE	70.0	0.69	42.1
German	Indo-European	whisper-large	EmoDB	86.0	0.85	55.5
Greek	Indo-European	hubert-base-ls960	AESDD	75.0	0.73	46.6
Hindi	Indo-European	whisper-small	VishalB	71.0	0.70	48.8
Hungarian	Uralic	hubert-base-ls960	HESD	52.0	0.52	36.7
Indonesian	Austronesian	hubert-base-ls960	IndoWaveSentiment	99.0	0.99	72.1
Italian	Indo-European	whisper-large	Emozionalmente	80.0	0.80	50.4
Japanese	Japonic	whisper-large	JVNV	87.0	0.87	61.9
Kannada	Dravidian	wavlm-large	Agrawal et al (2022)	48.0	0.44	27.5
Kazakh	Turkic	whisper-small	KazEmoTTS	65.0	0.64	44.0
Korean	Koreanic	whisper-small	KESDy18	81.0	0.81	57.7
Odia	Indo-European	wavlm-large	SITB-OSED	83.0	0.83	64.4
Persian	Indo-European	whisper-small	ShEMO	83.0	0.62	67.1
Polish	Indo-European	whisper-large	nEMO	87.0	0.87	56.6
Portuguese	Indo-European	whisper-small	emoUERJ	91.0	0.93	60.0
Quechua	Quechuan	whisper-large	Quechua Collao Corpus	83.0	0.83	58.7
Russian	Indo-European	hubert-base-ls960	RESD	51.0	0.51	34.5
Spanish	Indo-European	whisper-small	MESD	68.0	0.68	36.5
Swahili	Niger-Congo	wavlm-base-plus	Kenyan Swahili SER	83.0	0.80	61.5
Tamil	Dravidian	whisper-small	EmoTa	44.0	0.43	30.6
Telugu	Dravidian	hubert-base-ls960	Sowmith et al (2024)	53.0	0.52	40.5
Turkish	Turkic	whisper-small	TurEV-DB	84.0	0.84	54.9
Urdu	Indo-European	wav2vec2-base	URDU-Dataset	88.0	0.88	49.6

Dataset × model matrix

Test accuracy for every dataset and model. Model columns run from HuBERT through wav2vec 2.0 and WavLM to Whisper.

**Test accuracy, dataset × model.** Darker cells are higher. The last column is the mean over the 11 models.
Language	Dataset	hubert-base-ls960	hubert-large-ls960-ft	wav2vec2-base	wav2vec2-large-960h	wav2vec2-large-lv60	wav2vec2-xls-r-300m	wav2vec2-xls-r-1b	wavlm-base-plus	wavlm-large	whisper-small	whisper-large	Mean
Afrikaans	AfrikaansSE	0.67	0.41	0.66	0.43	0.40	0.37	0.24	0.59	0.63	0.71	0.65	0.52
Amharic	Amharic Speech Emotion Dataset (ASED)	0.89	0.63	0.75	0.32	0.27	0.32	0.40	0.84	0.82	0.90	0.86	0.64
Arabic	ANAD	0.58	0.66	0.57	0.65	0.64	0.61	0.63	0.61	0.71	0.79	0.82	0.66
Bengali	BANSpEmo	0.41	0.23	0.32	0.23	0.25	0.23	0.26	0.35	0.40	0.38	0.40	0.31
Bengali	SUBESCO	0.78	0.54	0.68	0.24	0.27	0.48	0.38	0.78	0.74	0.89	0.89	0.61
Chinese	ESD	0.92	0.73	0.90	0.45	0.28	0.59	0.68	0.90	0.91	0.97	0.98	0.76
English	IESC	0.82	0.53	0.66	0.31	0.25	0.25	0.24	0.78	0.81	0.71	0.74	0.55
English	RAVDESS	0.65	0.35	0.59	0.15	0.27	0.34	0.23	0.69	0.55	0.61	0.68	0.46
French	CaFE	0.60	0.33	0.38	0.24	0.21	0.25	0.25	0.53	0.70	0.61	0.53	0.42
German	EmoDB	0.78	0.44	0.70	0.34	0.26	0.35	0.30	0.70	0.53	0.84	0.86	0.55
Greek	AESDD	0.75	0.32	0.54	0.32	0.25	0.36	0.26	0.63	0.57	0.57	0.56	0.47
Hindi	VishalB	0.70	0.31	0.61	0.23	0.20	0.22	0.35	0.65	0.70	0.71	0.69	0.49
Hungarian	HESD	0.52	0.21	0.43	0.24	0.23	0.23	0.22	0.51	0.49	0.52	0.44	0.37
Indonesian	IndoWaveSentiment	0.99	0.46	0.77	0.37	0.57	0.46	0.59	0.94	0.94	0.97	0.87	0.72
Italian	EMOVO	0.57	0.28	0.56	0.28	0.26	0.25	0.24	0.58	0.55	0.64	0.70	0.45
Italian	Emozionalmente	0.71	0.49	0.61	0.32	0.30	0.47	0.28	0.72	0.73	0.75	0.80	0.56
Japanese	JVNV	0.86	0.49	0.67	0.39	0.26	0.43	0.34	0.84	0.85	0.81	0.87	0.62
Kannada	Agrawal et al (2022)	0.29	0.19	0.22	0.27	0.23	0.17	0.26	0.26	0.48	0.30	0.36	0.28
Kazakh	KazEmoTTS	0.51	0.29	0.53	0.24	0.21	0.35	0.32	0.51	0.58	0.65	0.65	0.44
Korean	KESDy18	0.80	0.46	0.70	0.23	0.24	0.39	0.43	0.74	0.76	0.81	0.79	0.58
Odia	SITB-OSED	0.75	0.62	0.64	0.43	0.28	0.51	0.58	0.78	0.83	0.83	0.83	0.64
Persian	ShEMO	0.80	0.58	0.76	0.44	0.43	0.46	0.67	0.79	0.80	0.83	0.82	0.67
Polish	nEMO	0.79	0.39	0.69	0.24	0.22	0.37	0.41	0.74	0.66	0.85	0.87	0.57
Portuguese	emoUERJ	0.74	0.51	0.63	0.23	0.29	0.47	0.44	0.74	0.76	0.91	0.88	0.60
Quechua	Quechua Collao Corpus	0.80	0.50	0.76	0.36	0.22	0.43	0.30	0.71	0.73	0.82	0.83	0.59
Russian	RESD	0.51	0.22	0.41	0.24	0.20	0.21	0.26	0.43	0.40	0.45	0.46	0.34
Spanish	MESD	0.58	0.26	0.27	0.26	0.13	0.24	0.24	0.28	0.41	0.68	0.67	0.37
Swahili	Kenyan Swahili SER	0.81	0.60	0.72	0.44	0.35	0.33	0.45	0.83	0.74	0.69	0.80	0.61
Tamil	EmoTa	0.38	0.25	0.38	0.24	0.19	0.24	0.20	0.35	0.29	0.44	0.41	0.31
Telugu	Sowmith et al (2024)	0.53	0.35	0.44	0.35	0.28	0.29	0.29	0.53	0.43	0.52	0.45	0.41
Turkish	TurEV-DB	0.68	0.44	0.48	0.40	0.39	0.34	0.48	0.52	0.63	0.84	0.84	0.55
Urdu	URDU-Dataset	0.87	0.43	0.88	0.34	0.35	0.41	0.35	0.76	0.50	0.78	0.82	0.59
Urdu	UrduSER	0.51	0.32	0.46	0.24	0.23	0.28	0.26	0.52	0.45	0.56	0.60	0.40
Mean over 33 datasets		0.68	0.42	0.59	0.32	0.29	0.35	0.36	0.64	0.64	0.71	0.71

All runs

Filter by language, model family or model, or search. Click a column header to sort.

363 runs

**All 363 runs.** Validation accuracy, test accuracy, and test macro-F1 and weighted-F1. Each run links to its classification reports, confusion matrices and loss curve on GitHub.
Language	Dataset	Model	Val acc.	Test acc.	Macro-F1	Weighted-F1	Logs
Afrikaans	AfrikaansSE	hubert-base-ls960	0.65	0.67	0.62	0.66	logs
Afrikaans	AfrikaansSE	hubert-large-ls960-ft	0.32	0.41	0.41	0.41	logs
Afrikaans	AfrikaansSE	wav2vec2-base	0.57	0.66	0.67	0.67	logs
Afrikaans	AfrikaansSE	wav2vec2-large-960h	0.38	0.43	0.41	0.41	logs
Afrikaans	AfrikaansSE	wav2vec2-large-lv60	0.36	0.40	0.40	0.40	logs
Afrikaans	AfrikaansSE	wav2vec2-xls-r-300m	0.38	0.37	0.37	0.37	logs
Afrikaans	AfrikaansSE	wav2vec2-xls-r-1b	0.35	0.24	0.15	0.11	logs
Afrikaans	AfrikaansSE	wavlm-base-plus	0.53	0.59	0.60	0.60	logs
Afrikaans	AfrikaansSE	wavlm-large	0.29	0.63	0.58	0.61	logs
Afrikaans	AfrikaansSE	whisper-small	0.75	0.71	0.70	0.70	logs
Afrikaans	AfrikaansSE	whisper-large	0.68	0.65	0.63	0.63	logs
Amharic	Amharic Speech Emotion Dataset (ASED)	hubert-base-ls960	0.88	0.89	0.89	0.89	logs
Amharic	Amharic Speech Emotion Dataset (ASED)	hubert-large-ls960-ft	0.64	0.63	0.63	0.63	logs
Amharic	Amharic Speech Emotion Dataset (ASED)	wav2vec2-base	0.75	0.75	0.75	0.75	logs
Amharic	Amharic Speech Emotion Dataset (ASED)	wav2vec2-large-960h	0.30	0.32	0.27	0.26	logs
Amharic	Amharic Speech Emotion Dataset (ASED)	wav2vec2-large-lv60	0.23	0.27	0.25	0.24	logs
Amharic	Amharic Speech Emotion Dataset (ASED)	wav2vec2-xls-r-300m	0.31	0.32	0.29	0.30	logs
Amharic	Amharic Speech Emotion Dataset (ASED)	wav2vec2-xls-r-1b	0.35	0.40	0.32	0.33	logs
Amharic	Amharic Speech Emotion Dataset (ASED)	wavlm-base-plus	0.87	0.84	0.84	0.84	logs
Amharic	Amharic Speech Emotion Dataset (ASED)	wavlm-large	0.80	0.82	0.82	0.82	logs
Amharic	Amharic Speech Emotion Dataset (ASED)	whisper-small	0.91	0.90	0.90	0.90	logs
Amharic	Amharic Speech Emotion Dataset (ASED)	whisper-large	0.88	0.86	0.86	0.86	logs
Arabic	ANAD	hubert-base-ls960	0.63	0.58	0.58	0.58	logs
Arabic	ANAD	hubert-large-ls960-ft	0.64	0.66	0.51	0.59	logs
Arabic	ANAD	wav2vec2-base	0.62	0.57	0.45	0.48	logs
Arabic	ANAD	wav2vec2-large-960h	0.62	0.65	0.54	0.61	logs
Arabic	ANAD	wav2vec2-large-lv60	0.64	0.64	0.39	0.51	logs
Arabic	ANAD	wav2vec2-xls-r-300m	0.61	0.61	0.46	0.54	logs
Arabic	ANAD	wav2vec2-xls-r-1b	0.61	0.63	0.47	0.56	logs
Arabic	ANAD	wavlm-base-plus	0.66	0.61	0.54	0.57	logs
Arabic	ANAD	wavlm-large	0.66	0.71	0.57	0.64	logs
Arabic	ANAD	whisper-small	0.72	0.79	0.78	0.79	logs
Arabic	ANAD	whisper-large	0.78	0.82	0.81	0.81	logs
Bengali	BANSpEmo	hubert-base-ls960	0.52	0.41	0.42	0.41	logs
Bengali	BANSpEmo	hubert-large-ls960-ft	0.24	0.23	0.22	0.22	logs
Bengali	BANSpEmo	wav2vec2-base	0.41	0.32	0.31	0.30	logs
Bengali	BANSpEmo	wav2vec2-large-960h	0.23	0.23	0.23	0.24	logs
Bengali	BANSpEmo	wav2vec2-large-lv60	0.26	0.25	0.25	0.26	logs
Bengali	BANSpEmo	wav2vec2-xls-r-300m	0.25	0.23	0.23	0.23	logs
Bengali	BANSpEmo	wav2vec2-xls-r-1b	0.19	0.26	0.25	0.25	logs
Bengali	BANSpEmo	wavlm-base-plus	0.49	0.35	0.36	0.35	logs
Bengali	BANSpEmo	wavlm-large	0.41	0.40	0.38	0.39	logs
Bengali	BANSpEmo	whisper-small	0.30	0.38	0.34	0.37	logs
Bengali	BANSpEmo	whisper-large	0.35	0.40	0.36	0.39	logs
Bengali	SUBESCO	hubert-base-ls960	0.77	0.78	0.78	0.78	logs
Bengali	SUBESCO	hubert-large-ls960-ft	0.56	0.54	0.53	0.52	logs
Bengali	SUBESCO	wav2vec2-base	0.69	0.68	0.68	0.68	logs
Bengali	SUBESCO	wav2vec2-large-960h	0.24	0.24	0.18	0.17	logs
Bengali	SUBESCO	wav2vec2-large-lv60	0.28	0.27	0.23	0.22	logs
Bengali	SUBESCO	wav2vec2-xls-r-300m	0.51	0.48	0.47	0.47	logs
Bengali	SUBESCO	wav2vec2-xls-r-1b	0.36	0.38	0.33	0.34	logs
Bengali	SUBESCO	wavlm-base-plus	0.76	0.78	0.77	0.77	logs
Bengali	SUBESCO	wavlm-large	0.76	0.74	0.74	0.74	logs
Bengali	SUBESCO	whisper-small	0.88	0.89	0.88	0.89	logs
Bengali	SUBESCO	whisper-large	0.88	0.89	0.89	0.89	logs
Chinese	ESD	hubert-base-ls960	0.93	0.92	0.91	0.91	logs
Chinese	ESD	hubert-large-ls960-ft	0.75	0.73	0.74	0.73	logs
Chinese	ESD	wav2vec2-base	0.91	0.90	0.90	0.90	logs
Chinese	ESD	wav2vec2-large-960h	0.45	0.45	0.42	0.42	logs
Chinese	ESD	wav2vec2-large-lv60	0.28	0.28	0.18	0.18	logs
Chinese	ESD	wav2vec2-xls-r-300m	0.62	0.59	0.58	0.58	logs
Chinese	ESD	wav2vec2-xls-r-1b	0.70	0.68	0.68	0.68	logs
Chinese	ESD	wavlm-base-plus	0.91	0.90	0.90	0.90	logs
Chinese	ESD	wavlm-large	0.90	0.91	0.91	0.91	logs
Chinese	ESD	whisper-small	0.97	0.97	0.97	0.97	logs
Chinese	ESD	whisper-large	0.98	0.98	0.98	0.98	logs
English	IESC	hubert-base-ls960	0.82	0.82	0.81	0.82	logs
English	IESC	hubert-large-ls960-ft	0.45	0.53	0.49	0.52	logs
English	IESC	wav2vec2-base	0.70	0.66	0.66	0.66	logs
English	IESC	wav2vec2-large-960h	0.28	0.31	0.27	0.28	logs
English	IESC	wav2vec2-large-lv60	0.21	0.25	0.20	0.22	logs
English	IESC	wav2vec2-xls-r-300m	0.26	0.25	0.24	0.25	logs
English	IESC	wav2vec2-xls-r-1b	0.28	0.24	0.22	0.22	logs
English	IESC	wavlm-base-plus	0.70	0.78	0.78	0.78	logs
English	IESC	wavlm-large	0.76	0.81	0.79	0.80	logs
English	IESC	whisper-small	0.78	0.71	0.69	0.71	logs
English	IESC	whisper-large	0.79	0.74	0.73	0.74	logs
English	RAVDESS	hubert-base-ls960	0.65	0.65	0.65	0.64	logs
English	RAVDESS	hubert-large-ls960-ft	0.34	0.35	0.34	0.33	logs
English	RAVDESS	wav2vec2-base	0.51	0.59	0.58	0.59	logs
English	RAVDESS	wav2vec2-large-960h	0.13	0.15	0.12	0.12	logs
English	RAVDESS	wav2vec2-large-lv60	0.23	0.27	0.21	0.25	logs
English	RAVDESS	wav2vec2-xls-r-300m	0.31	0.34	0.27	0.32	logs
English	RAVDESS	wav2vec2-xls-r-1b	0.22	0.23	0.18	0.21	logs
English	RAVDESS	wavlm-base-plus	0.68	0.69	0.69	0.69	logs
English	RAVDESS	wavlm-large	0.66	0.55	0.54	0.55	logs
English	RAVDESS	whisper-small	0.70	0.61	0.58	0.62	logs
English	RAVDESS	whisper-large	0.75	0.68	0.64	0.68	logs
French	CaFE	hubert-base-ls960	0.64	0.60	0.60	0.60	logs
French	CaFE	hubert-large-ls960-ft	0.33	0.33	0.32	0.33	logs
French	CaFE	wav2vec2-base	0.46	0.38	0.36	0.35	logs
French	CaFE	wav2vec2-large-960h	0.26	0.24	0.22	0.21	logs
French	CaFE	wav2vec2-large-lv60	0.20	0.21	0.17	0.15	logs
French	CaFE	wav2vec2-xls-r-300m	0.29	0.25	0.21	0.19	logs
French	CaFE	wav2vec2-xls-r-1b	0.26	0.25	0.19	0.18	logs
French	CaFE	wavlm-base-plus	0.53	0.53	0.52	0.53	logs
French	CaFE	wavlm-large	0.72	0.70	0.69	0.70	logs
French	CaFE	whisper-small	0.55	0.61	0.60	0.60	logs
French	CaFE	whisper-large	0.49	0.53	0.52	0.53	logs
German	EmoDB	hubert-base-ls960	0.75	0.78	0.76	0.78	logs
German	EmoDB	hubert-large-ls960-ft	0.43	0.44	0.39	0.38	logs
German	EmoDB	wav2vec2-base	0.68	0.70	0.68	0.70	logs
German	EmoDB	wav2vec2-large-960h	0.31	0.34	0.30	0.30	logs
German	EmoDB	wav2vec2-large-lv60	0.30	0.26	0.09	0.11	logs
German	EmoDB	wav2vec2-xls-r-300m	0.33	0.35	0.26	0.27	logs
German	EmoDB	wav2vec2-xls-r-1b	0.33	0.30	0.28	0.24	logs
German	EmoDB	wavlm-base-plus	0.71	0.70	0.64	0.67	logs
German	EmoDB	wavlm-large	0.52	0.53	0.44	0.44	logs
German	EmoDB	whisper-small	0.82	0.84	0.83	0.84	logs
German	EmoDB	whisper-large	0.84	0.86	0.85	0.86	logs
Greek	AESDD	hubert-base-ls960	0.69	0.75	0.73	0.74	logs
Greek	AESDD	hubert-large-ls960-ft	0.43	0.32	0.27	0.28	logs
Greek	AESDD	wav2vec2-base	0.58	0.54	0.53	0.55	logs
Greek	AESDD	wav2vec2-large-960h	0.37	0.32	0.22	0.24	logs
Greek	AESDD	wav2vec2-large-lv60	0.28	0.25	0.24	0.24	logs
Greek	AESDD	wav2vec2-xls-r-300m	0.38	0.36	0.28	0.30	logs
Greek	AESDD	wav2vec2-xls-r-1b	0.21	0.26	0.20	0.20	logs
Greek	AESDD	wavlm-base-plus	0.62	0.63	0.60	0.61	logs
Greek	AESDD	wavlm-large	0.70	0.57	0.52	0.53	logs
Greek	AESDD	whisper-small	0.60	0.57	0.56	0.58	logs
Greek	AESDD	whisper-large	0.57	0.56	0.56	0.57	logs
Hindi	VishalB	hubert-base-ls960	0.72	0.70	0.70	0.71	logs
Hindi	VishalB	hubert-large-ls960-ft	0.33	0.31	0.26	0.25	logs
Hindi	VishalB	wav2vec2-base	0.62	0.61	0.60	0.61	logs
Hindi	VishalB	wav2vec2-large-960h	0.22	0.23	0.13	0.12	logs
Hindi	VishalB	wav2vec2-large-lv60	0.20	0.20	0.12	0.11	logs
Hindi	VishalB	wav2vec2-xls-r-300m	0.22	0.22	0.14	0.13	logs
Hindi	VishalB	wav2vec2-xls-r-1b	0.35	0.35	0.34	0.34	logs
Hindi	VishalB	wavlm-base-plus	0.70	0.65	0.64	0.65	logs
Hindi	VishalB	wavlm-large	0.69	0.70	0.70	0.69	logs
Hindi	VishalB	whisper-small	0.68	0.71	0.70	0.70	logs
Hindi	VishalB	whisper-large	0.67	0.69	0.68	0.69	logs
Hungarian	HESD	hubert-base-ls960	0.54	0.52	0.52	0.51	logs
Hungarian	HESD	hubert-large-ls960-ft	0.29	0.21	0.19	0.18	logs
Hungarian	HESD	wav2vec2-base	0.42	0.43	0.40	0.41	logs
Hungarian	HESD	wav2vec2-large-960h	0.28	0.24	0.22	0.21	logs
Hungarian	HESD	wav2vec2-large-lv60	0.22	0.23	0.17	0.16	logs
Hungarian	HESD	wav2vec2-xls-r-300m	0.28	0.23	0.22	0.21	logs
Hungarian	HESD	wav2vec2-xls-r-1b	0.19	0.22	0.14	0.13	logs
Hungarian	HESD	wavlm-base-plus	0.52	0.51	0.50	0.49	logs
Hungarian	HESD	wavlm-large	0.45	0.49	0.48	0.47	logs
Hungarian	HESD	whisper-small	0.41	0.52	0.51	0.51	logs
Hungarian	HESD	whisper-large	0.48	0.44	0.43	0.43	logs
Indonesian	IndoWaveSentiment	hubert-base-ls960	0.97	0.99	0.99	0.99	logs
Indonesian	IndoWaveSentiment	hubert-large-ls960-ft	0.63	0.46	0.45	0.41	logs
Indonesian	IndoWaveSentiment	wav2vec2-base	0.78	0.77	0.76	0.76	logs
Indonesian	IndoWaveSentiment	wav2vec2-large-960h	0.61	0.37	0.32	0.26	logs
Indonesian	IndoWaveSentiment	wav2vec2-large-lv60	0.50	0.57	0.53	0.57	logs
Indonesian	IndoWaveSentiment	wav2vec2-xls-r-300m	0.58	0.46	0.46	0.45	logs
Indonesian	IndoWaveSentiment	wav2vec2-xls-r-1b	0.63	0.59	0.58	0.59	logs
Indonesian	IndoWaveSentiment	wavlm-base-plus	0.94	0.94	0.94	0.94	logs
Indonesian	IndoWaveSentiment	wavlm-large	0.97	0.94	0.94	0.95	logs
Indonesian	IndoWaveSentiment	whisper-small	0.96	0.97	0.97	0.98	logs
Indonesian	IndoWaveSentiment	whisper-large	0.85	0.87	0.86	0.87	logs
Italian	EMOVO	hubert-base-ls960	0.66	0.57	0.60	0.61	logs
Italian	EMOVO	hubert-large-ls960-ft	0.26	0.28	0.24	0.23	logs
Italian	EMOVO	wav2vec2-base	0.47	0.56	0.52	0.55	logs
Italian	EMOVO	wav2vec2-large-960h	0.26	0.28	0.20	0.20	logs
Italian	EMOVO	wav2vec2-large-lv60	0.21	0.26	0.18	0.18	logs
Italian	EMOVO	wav2vec2-xls-r-300m	0.25	0.25	0.22	0.21	logs
Italian	EMOVO	wav2vec2-xls-r-1b	0.31	0.24	0.23	0.22	logs
Italian	EMOVO	wavlm-base-plus	0.69	0.58	0.57	0.60	logs
Italian	EMOVO	wavlm-large	0.64	0.55	0.55	0.55	logs
Italian	EMOVO	whisper-small	0.78	0.64	0.64	0.64	logs
Italian	EMOVO	whisper-large	0.78	0.70	0.71	0.70	logs
Italian	Emozionalmente	hubert-base-ls960	0.72	0.71	0.71	0.71	logs
Italian	Emozionalmente	hubert-large-ls960-ft	0.46	0.49	0.46	0.46	logs
Italian	Emozionalmente	wav2vec2-base	0.62	0.61	0.61	0.61	logs
Italian	Emozionalmente	wav2vec2-large-960h	0.30	0.32	0.24	0.25	logs
Italian	Emozionalmente	wav2vec2-large-lv60	0.28	0.30	0.25	0.25	logs
Italian	Emozionalmente	wav2vec2-xls-r-300m	0.46	0.47	0.47	0.47	logs
Italian	Emozionalmente	wav2vec2-xls-r-1b	0.26	0.28	0.19	0.20	logs
Italian	Emozionalmente	wavlm-base-plus	0.75	0.72	0.72	0.72	logs
Italian	Emozionalmente	wavlm-large	0.69	0.73	0.73	0.73	logs
Italian	Emozionalmente	whisper-small	0.77	0.75	0.75	0.75	logs
Italian	Emozionalmente	whisper-large	0.79	0.80	0.80	0.80	logs
Japanese	JVNV	hubert-base-ls960	0.77	0.86	0.86	0.86	logs
Japanese	JVNV	hubert-large-ls960-ft	0.46	0.49	0.47	0.46	logs
Japanese	JVNV	wav2vec2-base	0.61	0.67	0.65	0.66	logs
Japanese	JVNV	wav2vec2-large-960h	0.33	0.39	0.30	0.29	logs
Japanese	JVNV	wav2vec2-large-lv60	0.31	0.26	0.22	0.22	logs
Japanese	JVNV	wav2vec2-xls-r-300m	0.42	0.43	0.40	0.40	logs
Japanese	JVNV	wav2vec2-xls-r-1b	0.40	0.34	0.31	0.29	logs
Japanese	JVNV	wavlm-base-plus	0.75	0.84	0.84	0.84	logs
Japanese	JVNV	wavlm-large	0.87	0.85	0.85	0.85	logs
Japanese	JVNV	whisper-small	0.80	0.81	0.81	0.81	logs
Japanese	JVNV	whisper-large	0.83	0.87	0.87	0.87	logs
Kannada	Agrawal et al (2022)	hubert-base-ls960	0.21	0.29	0.28	0.25	logs
Kannada	Agrawal et al (2022)	hubert-large-ls960-ft	0.19	0.19	0.16	0.17	logs
Kannada	Agrawal et al (2022)	wav2vec2-base	0.18	0.22	0.17	0.17	logs
Kannada	Agrawal et al (2022)	wav2vec2-large-960h	0.16	0.27	0.19	0.20	logs
Kannada	Agrawal et al (2022)	wav2vec2-large-lv60	0.18	0.23	0.19	0.21	logs
Kannada	Agrawal et al (2022)	wav2vec2-xls-r-300m	0.26	0.17	0.17	0.16	logs
Kannada	Agrawal et al (2022)	wav2vec2-xls-r-1b	0.15	0.26	0.20	0.25	logs
Kannada	Agrawal et al (2022)	wavlm-base-plus	0.19	0.26	0.20	0.16	logs
Kannada	Agrawal et al (2022)	wavlm-large	0.36	0.48	0.44	0.47	logs
Kannada	Agrawal et al (2022)	whisper-small	0.41	0.30	0.30	0.31	logs
Kannada	Agrawal et al (2022)	whisper-large	0.37	0.36	0.35	0.35	logs
Kazakh	KazEmoTTS	hubert-base-ls960	0.54	0.51	0.50	0.50	logs
Kazakh	KazEmoTTS	hubert-large-ls960-ft	0.28	0.29	0.25	0.25	logs
Kazakh	KazEmoTTS	wav2vec2-base	0.52	0.53	0.52	0.52	logs
Kazakh	KazEmoTTS	wav2vec2-large-960h	0.22	0.24	0.20	0.20	logs
Kazakh	KazEmoTTS	wav2vec2-large-lv60	0.21	0.21	0.14	0.14	logs
Kazakh	KazEmoTTS	wav2vec2-xls-r-300m	0.37	0.35	0.34	0.33	logs
Kazakh	KazEmoTTS	wav2vec2-xls-r-1b	0.30	0.32	0.30	0.30	logs
Kazakh	KazEmoTTS	wavlm-base-plus	0.52	0.51	0.49	0.50	logs
Kazakh	KazEmoTTS	wavlm-large	0.58	0.58	0.57	0.56	logs
Kazakh	KazEmoTTS	whisper-small	0.70	0.65	0.64	0.65	logs
Kazakh	KazEmoTTS	whisper-large	0.70	0.65	0.63	0.64	logs
Korean	KESDy18	hubert-base-ls960	0.76	0.80	0.79	0.80	logs
Korean	KESDy18	hubert-large-ls960-ft	0.46	0.46	0.45	0.45	logs
Korean	KESDy18	wav2vec2-base	0.67	0.70	0.68	0.69	logs
Korean	KESDy18	wav2vec2-large-960h	0.25	0.23	0.14	0.12	logs
Korean	KESDy18	wav2vec2-large-lv60	0.25	0.24	0.20	0.19	logs
Korean	KESDy18	wav2vec2-xls-r-300m	0.40	0.39	0.38	0.38	logs
Korean	KESDy18	wav2vec2-xls-r-1b	0.43	0.43	0.42	0.42	logs
Korean	KESDy18	wavlm-base-plus	0.70	0.74	0.72	0.74	logs
Korean	KESDy18	wavlm-large	0.77	0.76	0.75	0.76	logs
Korean	KESDy18	whisper-small	0.81	0.81	0.81	0.81	logs
Korean	KESDy18	whisper-large	0.78	0.79	0.79	0.79	logs
Odia	SITB-OSED	hubert-base-ls960	0.76	0.75	0.74	0.75	logs
Odia	SITB-OSED	hubert-large-ls960-ft	0.61	0.62	0.62	0.62	logs
Odia	SITB-OSED	wav2vec2-base	0.65	0.64	0.60	0.62	logs
Odia	SITB-OSED	wav2vec2-large-960h	0.44	0.43	0.39	0.40	logs
Odia	SITB-OSED	wav2vec2-large-lv60	0.29	0.28	0.20	0.21	logs
Odia	SITB-OSED	wav2vec2-xls-r-300m	0.50	0.51	0.48	0.49	logs
Odia	SITB-OSED	wav2vec2-xls-r-1b	0.54	0.58	0.53	0.55	logs
Odia	SITB-OSED	wavlm-base-plus	0.79	0.78	0.77	0.77	logs
Odia	SITB-OSED	wavlm-large	0.82	0.83	0.83	0.83	logs
Odia	SITB-OSED	whisper-small	0.83	0.83	0.83	0.83	logs
Odia	SITB-OSED	whisper-large	0.84	0.83	0.83	0.83	logs
Persian	ShEMO	hubert-base-ls960	0.77	0.80	0.54	0.78	logs
Persian	ShEMO	hubert-large-ls960-ft	0.57	0.58	0.31	0.53	logs
Persian	ShEMO	wav2vec2-base	0.71	0.76	0.46	0.73	logs
Persian	ShEMO	wav2vec2-large-960h	0.44	0.44	0.18	0.33	logs
Persian	ShEMO	wav2vec2-large-lv60	0.43	0.43	0.16	0.30	logs
Persian	ShEMO	wav2vec2-xls-r-300m	0.44	0.46	0.19	0.35	logs
Persian	ShEMO	wav2vec2-xls-r-1b	0.60	0.67	0.36	0.63	logs
Persian	ShEMO	wavlm-base-plus	0.77	0.79	0.52	0.77	logs
Persian	ShEMO	wavlm-large	0.72	0.80	0.59	0.79	logs
Persian	ShEMO	whisper-small	0.79	0.83	0.62	0.82	logs
Persian	ShEMO	whisper-large	0.80	0.82	0.61	0.81	logs
Polish	nEMO	hubert-base-ls960	0.77	0.79	0.79	0.79	logs
Polish	nEMO	hubert-large-ls960-ft	0.40	0.39	0.37	0.37	logs
Polish	nEMO	wav2vec2-base	0.67	0.69	0.68	0.68	logs
Polish	nEMO	wav2vec2-large-960h	0.24	0.24	0.16	0.16	logs
Polish	nEMO	wav2vec2-large-lv60	0.23	0.22	0.15	0.15	logs
Polish	nEMO	wav2vec2-xls-r-300m	0.36	0.37	0.36	0.36	logs
Polish	nEMO	wav2vec2-xls-r-1b	0.43	0.41	0.40	0.40	logs
Polish	nEMO	wavlm-base-plus	0.73	0.74	0.73	0.74	logs
Polish	nEMO	wavlm-large	0.66	0.66	0.67	0.67	logs
Polish	nEMO	whisper-small	0.85	0.85	0.85	0.85	logs
Polish	nEMO	whisper-large	0.88	0.87	0.87	0.87	logs
Portuguese	emoUERJ	hubert-base-ls960	0.91	0.74	0.72	0.74	logs
Portuguese	emoUERJ	hubert-large-ls960-ft	0.59	0.51	0.50	0.51	logs
Portuguese	emoUERJ	wav2vec2-base	0.70	0.63	0.56	0.60	logs
Portuguese	emoUERJ	wav2vec2-large-960h	0.41	0.23	0.23	0.24	logs
Portuguese	emoUERJ	wav2vec2-large-lv60	0.20	0.29	0.21	0.25	logs
Portuguese	emoUERJ	wav2vec2-xls-r-300m	0.42	0.47	0.43	0.47	logs
Portuguese	emoUERJ	wav2vec2-xls-r-1b	0.49	0.44	0.43	0.43	logs
Portuguese	emoUERJ	wavlm-base-plus	0.88	0.74	0.72	0.74	logs
Portuguese	emoUERJ	wavlm-large	0.88	0.76	0.75	0.77	logs
Portuguese	emoUERJ	whisper-small	0.74	0.91	0.93	0.91	logs
Portuguese	emoUERJ	whisper-large	0.73	0.88	0.87	0.88	logs
Quechua	Quechua Collao Corpus	hubert-base-ls960	0.81	0.80	0.79	0.80	logs
Quechua	Quechua Collao Corpus	hubert-large-ls960-ft	0.50	0.50	0.47	0.47	logs
Quechua	Quechua Collao Corpus	wav2vec2-base	0.78	0.76	0.76	0.76	logs
Quechua	Quechua Collao Corpus	wav2vec2-large-960h	0.39	0.36	0.33	0.33	logs
Quechua	Quechua Collao Corpus	wav2vec2-large-lv60	0.22	0.22	0.18	0.18	logs
Quechua	Quechua Collao Corpus	wav2vec2-xls-r-300m	0.43	0.43	0.43	0.43	logs
Quechua	Quechua Collao Corpus	wav2vec2-xls-r-1b	0.28	0.30	0.23	0.23	logs
Quechua	Quechua Collao Corpus	wavlm-base-plus	0.75	0.71	0.70	0.71	logs
Quechua	Quechua Collao Corpus	wavlm-large	0.76	0.73	0.73	0.73	logs
Quechua	Quechua Collao Corpus	whisper-small	0.82	0.82	0.81	0.82	logs
Quechua	Quechua Collao Corpus	whisper-large	0.84	0.83	0.83	0.83	logs
Russian	RESD	hubert-base-ls960	0.48	0.51	0.51	0.51	logs
Russian	RESD	hubert-large-ls960-ft	0.26	0.22	0.19	0.19	logs
Russian	RESD	wav2vec2-base	0.40	0.41	0.40	0.41	logs
Russian	RESD	wav2vec2-large-960h	0.23	0.24	0.21	0.22	logs
Russian	RESD	wav2vec2-large-lv60	0.20	0.20	0.17	0.18	logs
Russian	RESD	wav2vec2-xls-r-300m	0.22	0.21	0.18	0.19	logs
Russian	RESD	wav2vec2-xls-r-1b	0.29	0.26	0.26	0.25	logs
Russian	RESD	wavlm-base-plus	0.46	0.43	0.41	0.41	logs
Russian	RESD	wavlm-large	0.43	0.40	0.38	0.36	logs
Russian	RESD	whisper-small	0.49	0.45	0.45	0.45	logs
Russian	RESD	whisper-large	0.48	0.46	0.43	0.43	logs
Spanish	MESD	hubert-base-ls960	0.63	0.58	0.56	0.58	logs
Spanish	MESD	hubert-large-ls960-ft	0.33	0.26	0.23	0.24	logs
Spanish	MESD	wav2vec2-base	0.38	0.27	0.26	0.26	logs
Spanish	MESD	wav2vec2-large-960h	0.32	0.26	0.23	0.24	logs
Spanish	MESD	wav2vec2-large-lv60	0.15	0.13	0.10	0.09	logs
Spanish	MESD	wav2vec2-xls-r-300m	0.29	0.24	0.14	0.15	logs
Spanish	MESD	wav2vec2-xls-r-1b	0.27	0.24	0.22	0.21	logs
Spanish	MESD	wavlm-base-plus	0.40	0.28	0.27	0.28	logs
Spanish	MESD	wavlm-large	0.51	0.41	0.38	0.38	logs
Spanish	MESD	whisper-small	0.67	0.68	0.68	0.67	logs
Spanish	MESD	whisper-large	0.69	0.67	0.67	0.66	logs
Swahili	Kenyan Swahili SER	hubert-base-ls960	0.75	0.81	0.78	0.81	logs
Swahili	Kenyan Swahili SER	hubert-large-ls960-ft	0.61	0.60	0.59	0.59	logs
Swahili	Kenyan Swahili SER	wav2vec2-base	0.65	0.72	0.69	0.71	logs
Swahili	Kenyan Swahili SER	wav2vec2-large-960h	0.37	0.44	0.43	0.44	logs
Swahili	Kenyan Swahili SER	wav2vec2-large-lv60	0.41	0.35	0.26	0.24	logs
Swahili	Kenyan Swahili SER	wav2vec2-xls-r-300m	0.35	0.33	0.33	0.33	logs
Swahili	Kenyan Swahili SER	wav2vec2-xls-r-1b	0.35	0.45	0.41	0.42	logs
Swahili	Kenyan Swahili SER	wavlm-base-plus	0.80	0.83	0.80	0.83	logs
Swahili	Kenyan Swahili SER	wavlm-large	0.78	0.74	0.74	0.74	logs
Swahili	Kenyan Swahili SER	whisper-small	0.71	0.69	0.69	0.69	logs
Swahili	Kenyan Swahili SER	whisper-large	0.78	0.80	0.79	0.79	logs
Tamil	EmoTa	hubert-base-ls960	0.42	0.38	0.38	0.38	logs
Tamil	EmoTa	hubert-large-ls960-ft	0.26	0.25	0.18	0.22	logs
Tamil	EmoTa	wav2vec2-base	0.39	0.38	0.37	0.37	logs
Tamil	EmoTa	wav2vec2-large-960h	0.30	0.24	0.15	0.18	logs
Tamil	EmoTa	wav2vec2-large-lv60	0.22	0.19	0.15	0.17	logs
Tamil	EmoTa	wav2vec2-xls-r-300m	0.19	0.24	0.17	0.19	logs
Tamil	EmoTa	wav2vec2-xls-r-1b	0.23	0.20	0.14	0.14	logs
Tamil	EmoTa	wavlm-base-plus	0.33	0.35	0.35	0.35	logs
Tamil	EmoTa	wavlm-large	0.33	0.29	0.27	0.27	logs
Tamil	EmoTa	whisper-small	0.40	0.44	0.43	0.43	logs
Tamil	EmoTa	whisper-large	0.40	0.41	0.40	0.40	logs
Telugu	Sowmith et al (2024)	hubert-base-ls960	0.48	0.53	0.52	0.53	logs
Telugu	Sowmith et al (2024)	hubert-large-ls960-ft	0.41	0.35	0.30	0.30	logs
Telugu	Sowmith et al (2024)	wav2vec2-base	0.46	0.44	0.42	0.43	logs
Telugu	Sowmith et al (2024)	wav2vec2-large-960h	0.28	0.35	0.27	0.27	logs
Telugu	Sowmith et al (2024)	wav2vec2-large-lv60	0.22	0.28	0.20	0.21	logs
Telugu	Sowmith et al (2024)	wav2vec2-xls-r-300m	0.28	0.29	0.24	0.25	logs
Telugu	Sowmith et al (2024)	wav2vec2-xls-r-1b	0.19	0.29	0.24	0.26	logs
Telugu	Sowmith et al (2024)	wavlm-base-plus	0.44	0.53	0.50	0.52	logs
Telugu	Sowmith et al (2024)	wavlm-large	0.46	0.43	0.40	0.39	logs
Telugu	Sowmith et al (2024)	whisper-small	0.50	0.52	0.51	0.52	logs
Telugu	Sowmith et al (2024)	whisper-large	0.46	0.45	0.43	0.45	logs
Turkish	TurEV-DB	hubert-base-ls960	0.66	0.68	0.66	0.67	logs
Turkish	TurEV-DB	hubert-large-ls960-ft	0.44	0.44	0.34	0.37	logs
Turkish	TurEV-DB	wav2vec2-base	0.43	0.48	0.41	0.43	logs
Turkish	TurEV-DB	wav2vec2-large-960h	0.36	0.40	0.29	0.31	logs
Turkish	TurEV-DB	wav2vec2-large-lv60	0.37	0.39	0.27	0.30	logs
Turkish	TurEV-DB	wav2vec2-xls-r-300m	0.39	0.34	0.25	0.26	logs
Turkish	TurEV-DB	wav2vec2-xls-r-1b	0.47	0.48	0.44	0.45	logs
Turkish	TurEV-DB	wavlm-base-plus	0.58	0.52	0.51	0.51	logs
Turkish	TurEV-DB	wavlm-large	0.65	0.63	0.55	0.57	logs
Turkish	TurEV-DB	whisper-small	0.88	0.84	0.84	0.84	logs
Turkish	TurEV-DB	whisper-large	0.87	0.84	0.84	0.84	logs
Urdu	URDU-Dataset	hubert-base-ls960	0.79	0.87	0.86	0.87	logs
Urdu	URDU-Dataset	hubert-large-ls960-ft	0.35	0.43	0.38	0.42	logs
Urdu	URDU-Dataset	wav2vec2-base	0.72	0.88	0.88	0.87	logs
Urdu	URDU-Dataset	wav2vec2-large-960h	0.26	0.34	0.25	0.30	logs
Urdu	URDU-Dataset	wav2vec2-large-lv60	0.18	0.35	0.27	0.30	logs
Urdu	URDU-Dataset	wav2vec2-xls-r-300m	0.29	0.41	0.36	0.38	logs
Urdu	URDU-Dataset	wav2vec2-xls-r-1b	0.23	0.35	0.26	0.32	logs
Urdu	URDU-Dataset	wavlm-base-plus	0.72	0.76	0.75	0.76	logs
Urdu	URDU-Dataset	wavlm-large	0.64	0.50	0.50	0.49	logs
Urdu	URDU-Dataset	whisper-small	0.87	0.78	0.76	0.78	logs
Urdu	URDU-Dataset	whisper-large	0.85	0.82	0.79	0.83	logs
Urdu	UrduSER	hubert-base-ls960	0.53	0.51	0.50	0.51	logs
Urdu	UrduSER	hubert-large-ls960-ft	0.31	0.32	0.27	0.26	logs
Urdu	UrduSER	wav2vec2-base	0.44	0.46	0.44	0.45	logs
Urdu	UrduSER	wav2vec2-large-960h	0.23	0.24	0.19	0.18	logs
Urdu	UrduSER	wav2vec2-large-lv60	0.22	0.23	0.16	0.15	logs
Urdu	UrduSER	wav2vec2-xls-r-300m	0.27	0.28	0.27	0.27	logs
Urdu	UrduSER	wav2vec2-xls-r-1b	0.28	0.26	0.19	0.19	logs
Urdu	UrduSER	wavlm-base-plus	0.52	0.52	0.49	0.51	logs
Urdu	UrduSER	wavlm-large	0.44	0.45	0.44	0.44	logs
Urdu	UrduSER	whisper-small	0.59	0.56	0.55	0.56	logs
Urdu	UrduSER	whisper-large	0.60	0.60	0.59	0.60	logs

Dataset names match the catalog. Run IDs follow the folder names in results/train_val_test_logs/ (<iso>_<dataset>_<model>) and are also in runs.csv. If you use these numbers, please cite the paper.