Dolphin whistles are essential for social communication, and controllable synthesis of such signals is critical to studying behavioral mechanisms and enabling human-designed interaction stimuli. We propose a diffusion-based framework for dolphin whistle generation conditioned on a user-defined fundamental frequency (F0) contour sketch. This formulation allows humans to directly specify interpretable acoustic structures, which are transformed into realistic time–frequency spectra by the generative model. Compared with previous bioacoustic synthesis approaches, the proposed method provides explicit control over whistle semantics while preserving spectral fidelity. We further evaluate its utility for data augmentation in a CRNN9-based whistle classification system, where the generated samples improve robustness over conventional augmentation strategies. The proposed approach establishes a new paradigm for controllable bioacoustic signal generation, with potential applications in animal communication analysis and human–animal interaction studies.