Voice set off detection is a crucial activity, which allows activating a voice assistant when a goal consumer speaks a key phrase phrase. A detector is often educated on speech information impartial of speaker info and used for the voice set off detection activity. Nonetheless, such a speaker impartial voice set off detector usually suffers from efficiency degradation on speech from underrepresented teams, similar to accented audio system. On this work, we suggest a novel voice set off detector that may use a small variety of utterances from a goal speaker to enhance detection accuracy. Our proposed mannequin employs an encoder-decoder structure. Whereas the encoder performs speaker impartial voice set off detection, much like the standard detector, the decoder predicts a personalised embedding for every utterance. A customized voice set off rating is then obtained as a similarity rating between the embeddings of enrollment utterances and a take a look at utterance. The personalised embedding permits adapting to focus on speaker’s speech when computing the voice set off rating, therefore bettering voice set off detection accuracy. Experimental outcomes present that the proposed method achieves a 38% relative discount in a false rejection fee (FRR) in comparison with a baseline speaker impartial voice set off mannequin.