Multimodal Role-playing Speech Generation on SpeechRole 4 benchmark leaderboard · SOTA2 Research

Multimodal Role-playing Speech Generation on SpeechRole 4

226TTFA (ms)LLaMA

Evaluation Results

Method	Links
LLaMA2026.06		226	80	39.7	2.21	78.08	87.95	86.07	78.64	66.2	64.27	64.15	70.77	74.52
Qwen2.52026.06		274	80	45.3	0.98	51.27	67.14	63.26	64.74	47.63	47.93	39.31	59.07	55.04
GPT-4o2026.06		320	80	50.1	2.03	91.37	93.29	99.83	90.79	80.46	83.41	80.18	89.1	88.62
SpeechRole2026.06		389	80	43.3	5.31	82.03	87.45	93.16	78.38	65.05	69.13	61.22	83.34	77.47
DeSRPA (w/o Both CVs)2026.06		561	91.9	53.7	2.49	87.45	89.22	95.48	81.55	69.36	68.74	71.67	78.28	80.22
DeSRPA (w/o Speech CV)2026.06		569	89.2	54.9	2.52	87.99	88.46	93.51	81.49	71.86	72.45	75.92	80.74	81.68
DeSRPA (w/o LLM CV)2026.06		573	90.5	67.7	2.64	88.1	88.91	92.45	81.52	74.12	75.68	72.35	77.43	81.2
DeSRPA2026.06		577	88.6	70.1	2.63	87.9	87.41	95.06	81.47	79.58	81.6	76.15	81.16	83.79
AliCloud2026.06		872	85.9	69.4	1.74	89.86	87.06	95.63	81.77	77.43	78.72	74.02	84.05	83.56