Whisper is one of the current state-of-the-art systems for automatic speech recognition (ASR), both because of availability and performance. Recent previous research has shown that such systems may also be applied to more challenging domains like radio frequency (RF) speech processing, which typically faces the additional problem of locating speech within a time-frequency plane and estimating the in advance unknown carrier frequency (CF) and sideband (SB) of the modulated signals. This paper proposes a novel approach for detecting potentially several speech signals in a wideband RF signal, providing exact time-frequency positions and speechto- text (S2T) information. We combine a modified You Only Look Once (YOLO) for the voice activity detection (VAD) part with Whisper for the S2T part. The paper in particular analyzes the VAD performance and the influence of inaccurate CF estimates with respect to the background noise level and provides a first step towards an end-to-end ASR pipeline.