摘要 |
<p>A method for building the speech and/or language models (41, 42) which are initially loaded in a multimodal handheld device (4), such as a PDA, a mobile phone or a toy. The speech and language models are used for speech recognition. The method comprises the following steps: said speaker selects (406) on a visual user interface of said multimodal handheld device one or several speaker clusters to which said speaker belongs, said multimodal handheld device (4) sends (408) said selected clusters to a remote server (1), cluster-dependent speech and/or language models (41, 42) are downloaded (412) from said remote server (1) to said multimodal handheld device (4). <IMAGE></p> |