| 刊名 | Meteorological and Environmental Research |
| 作者 | Riyuan HAO, Kun MA*, Jianyong LIU, Xue HU, Haijiang ZHAO, Xinglu LI |
| 作者单位 | Zhangjiakou Meteorological Bureau, Zhangjiakou 075000, China |
| DOI | 10.19547/j.issn2152-3940.2026.04.04 |
| 年份 | 2026 |
| 刊期 | 4 |
| 页码 | 22-25 |
| 关键词 | Meteorological new media; Speech cloning; Message parsing; Volcanic engine; Index-TTS 2.0 |
| 摘要 | In response to practical issues such as equipment dependence, personnel scheduling risks, and disaster warning timeliness in grassroots meteorological program production, a meteorological voice broadcast auxiliary process of online and offline dual model switching is proposed in this paper. The process integrates three core functions: message parsing, model switching, and speech generation. At the same time, it is connected to the cloud volcano engine speech replication large model and the local Index-TS 2.0 open source model. Operators can manually switch as needed. The system is equipped with an updatable mandatory dictionary of high-frequency meteorological words to improve the accuracy of pronunciation of terms and place names. Subsequent tests have shown that the system can serve as a backup solution to avoid downtime in the event of equipment failure or broadcaster absence, compress the production cycle of sudden warnings to the minute level, and support mass production of multiple versions of programs. This paper provides a solution for emergency voice program synthesis for meteorological new media by complementing cloud high precision and local high reliability. |
