This paper summarises the Cambridge team’s work in the DASR Task of the CHiME-7 Challenge for speaker diarisation and automatic speech recognition (ASR). For speaker diarisation, the combination of Pyannote and ECAPA-TDNN was explored. For ASR, a two-pass ASR system was built. The first-pass model was based on a CTC model fine-tuned from a pre-trained WavLM model, based on which test-time unsupervised adaptation was implemented before decoded with a 4-gram language model (LM). For the second-pass system, with WavLM-based encoders, forward and backward hybrid CTC/attention models, as well as a label-synchronous neural transducer model, were trained for re-scoring. As a result of these efforts, for the sub-track, the Cambridge system achieved 21.7% and 22.7% DA-WER on the overall Dev and Eval sets respectively, with 24.6% and 32.0% relative error-rate reductions over the challenge baselines. For the main track, an ECAPA-based system was used for diarisation. Using our diarisation, together with our proposed ASR, the submitted main-track system gave a DA-WER of 38.7% on the Eval set which is with a 30% relative reduction in error rate compared to the challenge baseline.