Speaker recognition is a technology that identifies the speaker in an input utterance by extracting speaker-distinguishable features from the speech signal. Speaker recognition is used for system security and authentication; therefore, it is crucial to extract unique features of the speaker to achieve high recognition rates. Representative methods for extracting these features include a classification approach, or utilizing contrastive learning to learn the speaker relationship between representations and then using embeddings extracted from a specific layer of the model. This paper introduces a framework for developing robust speaker recognition models through contrastive learning. This approach aims to minimize the similarity to hard negative samples—those that are genuine negatives, but have extremely similar features to the positives, leading to potential mistaken. Specifically, our proposed method trains the model by estimating hard negative samples within a mini-batch during contrastive learning, and then utilizes a cross-attention mechanism to determine speaker agreement for pairs of utterances. To demonstrate the effectiveness of our proposed method, we compared the performance of a deep learning model trained with a conventional loss function utilized in speaker recognition with that of a deep learning model trained using our proposed method, as measured by the equal error rate (EER), an objective performance metric. Our results indicate that when trained with the voxceleb2 dataset, the proposed method achieved an EER of 0.98% on the voxceleb1-E dataset and 1.84% on the voxceleb1-H dataset.

Recent advances in neural text-to-speech (TTS) systems have achieved high-quality speech synthesis, but maintaining naturalness while controlling utterance duration remains a challenging task. This challenge is further complicated by the need to balance naturalness with the ability to control utterance duration, which is essential for creating realistic speech outputs. Existing approaches to controlling utterance duration often rely on post-processing techniques, which can compromise naturalness. Moreover, these techniques may not be effective in all scenarios, particularly when dealing with variable utterance durations. In this paper, we propose a novel masked duration model, enabling controllable utterance duration in TTS synthesis. Our approach utilizes an audio prompt, a text prompt, and a mask to predict phoneme durations within the masked span, which corresponds to the utterance duration. This allows us to precisely control the utterance duration by determining the target duration initially and then predicting the phoneme duration. Furthermore, our model enables fine-grained control over utterance duration, allowing for more nuanced and realistic speech outputs. Additionally, we employ adversarial training to enhance the robustness of alignment between the audio and text prompts. Experimental results demonstrate that our proposed model outperforms the baseline in terms of utterance duration control. Ablation studies validate the effectiveness of the adversarial training in improving model performance. This technology has significant implications for applications requiring precise control over utterance duration, such as automatic voice dubbing.

Recent advances in speech deepfake detection (SDD) have significantly improved artifacts-based detection in spoofed speech. However, most models overlook speech naturalness, a crucial cue for distinguishing bona fide speech from spoofed speech. This study proposes naturalness-aware curriculum learning, a novel training framework that leverages speech naturalness to enhance the robustness and generalization of SDD. This approach measures sample difficulty using both ground-truth labels and mean opinion scores, and adjusts the training schedule to progressively introduce more challenging samples. To further improve generalization, a dynamic temperature scaling method based on speech naturalness is incorporated into the training process. A 23% relative reduction in the EER was achieved in the experiments on the ASVspoof 2021 DF dataset, without modifying the model architecture. Ablation studies confirmed the effectiveness of naturalness-aware training strategies for SDD tasks.