The recent advancements in speech-to-speech translation (S2ST) and speech deepfake detection have shown significant progress, particularly in the development of direct S2ST models and the use of pre-trained self-supervised models for deepfake detection. Direct S2ST models are increasingly being explored for their potential to translate speech without intermediate text generation, offering advantages in decoding latency and feature preservation. However, these models still face challenges in achieving high-quality performance comparable to cascade models. On the other hand, the integration of pre-trained ASR models into deepfake detection systems has demonstrated promising results, highlighting a potential correlation between ASR performance and deepfake detection accuracy. Additionally, the use of synthetic interleaved data for scaling speech-text pre-training has enabled the development of more scalable and efficient speech language models, advancing the field of spoken language translation. These developments collectively indicate a shift towards more integrated and efficient solutions in speech processing, with a focus on leveraging existing models and synthetic data to enhance performance and scalability.