SwanVoice: Expressive Long-Form Zero-Shot Speech Synthesis for Both Monologue and Dialogue

Zero-shot text-to-speech (TTS) has improved substantially for single-speaker synthesis, yet expressive long-form multi-speaker dialogue remains difficult. A common workaround is to synthesize each turn with a monologue TTS model and stitch the outputs together. This adds inference cost and often breaks acoustic consistency, conversational coherence, and affective continuity across turns. Recent dialogue TTS systems have begun to address this setting, but they still struggle to keep expressive coherence, controllable speaker switching, and monologue quality at the same time. We present SwanData-Speech and SwanVoice. SwanData-Speech builds monologue and dialogue corpora from in-the-wild audio, using Swan Forced Aligner for pause-aware word-level alignment and RobustMegaTTS3 for pronunciation-hard cases. Built on these data, SwanVoice is a zero-shot TTS model for 1--4 speakers, combining a 25 Hz VAE, raw-text conditioning with pause-aware symbols and pinyin substitution, and a flow-matching DiT with speaker-turn conditioning. Training starts from monologue speech, moves through mixed and real dialogue data, and then uses DiffusionNFT post-training with phone-level and speaker-similarity rewards. On SwanBench-Speech, SwanVoice obtains higher richness and hierarchy scores than all evaluated open-source baselines in both monologue and dialogue settings, while content accuracy remains the main limitation. Audio demos are available at https://swanaigc.github.io//#swanvoice.

Towards Streaming Synchronized Spatial Audio Generation via Autoregressive Diffusion Transformer

Real-time and accurate spatial audio generation is pivotal for delivering an immersive experience. However, existing spatial audio synthesis technologies are often encumbered by a tradeoff between generation quality and high inference latency, as well as difficulty in capturing precise spatial information from multimodal inputs. To address these challenges, we propose SwanSphere, a unified streaming framework for high-fidelity spatial audio generation from panoramic videos and text prompts. SwanSphere mainly makes the following contributions: 1) We introduce a causal autoregressive diffusion transformer architecture that enables streaming high-quality spatial audio generation. 2) We design a Spatial Video-Audio Contrastive (SVAC) learning strategy to align the video encoder with the acoustic domain, and further employ a multi-objective online direct preference optimization (ODPO) scheme, resulting in strong spatial perception and robust multimodal spatial audio synthesis. 3) To alleviate the current scarcity of spatial audio datasets, we also develop an automated annotation pipeline for generating detailed spatial captions. Experimental results demonstrate that SwanSphere achieves superior performance in both video-to-spatial and text-to-spatial audio generation tasks. Demos can be found at: https://swanaigc.github.io.

Comprehensive Benchmarking of Long-Form Speech Generation in Diverse Scenarios

Recent advances in speech generation have enabled high-fidelity synthesis, yet systematic evaluation of models under long-context conditions remains largely underexplored. A comprehensive evaluation benchmark for long-form speech is indispensable for two reasons: 1) existing test scenarios are often confined to limited domains, creating a significant gap with the diverse downstream applications; 2) existing metrics overlook critical long-text factors such as consistency and coherence, failing to generalize reliably. To this end, we propose Swanbench-Speech, a comprehensive benchmark that decomposes long-form speech quality into specific, disentangled dimensions. SwanBench-Speech has three key properties. 1) Rich speech scenarios: Focusing on long-form speech generation and dialog generation, SwanBench-Speech covers acoustics, semantics, and expressiveness challenges, and consists of 1,101 samples spanning 17 common speech scenarios; 2) Comprehensive evaluation dimensions: Along the acoustics, semantics, and expressiveness axes, SwanBench-Speech defines an automated evaluation protocol with seven metrics to provide a comprehensive, accurate, and standardized assessment; 3) Valuable Insights: Through extensive experiments, we reveal that current models still struggle in highly expressive scenarios and exhibit a notable gap in consistency and hierarchy compared to real recordings.

T2A-Feedback: Improving Basic Capabilities of Text-to-Audio Generation via Fine-grained AI Feedback

Text-to-audio (T2A) generation has achieved remarkable progress in generating a variety of audio outputs from language prompts. However, current state-of-the-art T2A models still struggle to satisfy human preferences for prompt-following and acoustic quality when generating complex multi-event audio. To improve the performance of the model in these high-level applications, we propose to enhance the basic capabilities of the model with AI feedback learning. First, we introduce fine-grained AI audio scoring pipelines to: 1) verify whether each event in the text prompt is present in the audio (Event Occurrence Score), 2) detect deviations in event sequences from the language description (Event Sequence Score), and 3) assess the overall acoustic and harmonic quality of the generated audio (Acoustic&Harmonic Quality). We evaluate these three automatic scoring pipelines and find that they correlate significantly better with human preferences than other evaluation metrics. This highlights their value as both feedback signals and evaluation metrics. Utilizing our robust scoring pipelines, we construct a large audio preference dataset, T2A-FeedBack, which contains 41k prompts and 249k audios, each accompanied by detailed scores. Moreover, we introduce T2A-EpicBench, a benchmark that focuses on long captions, multi-events, and story-telling scenarios, aiming to evaluate the advanced capabilities of T2A models. Finally, we demonstrate how T2A-FeedBack can enhance current state-of-the-art audio model. With simple preference tuning, the audio generation model exhibits significant improvements in both simple (AudioCaps test set) and complex (T2A-EpicBench) scenarios.

Automated MRI Field of View Prescription from Region of Interest Prediction by Intra-stack Attention Neural Network

Manual prescription of the field of view (FOV) by MRI technologists is variable and prolongs the scanning process. Often, the FOV is too large or crops critical anatomy. We propose a deep-learning framework, trained by radiologists' supervision, for automating FOV prescription. An intra-stack shared feature extraction network and an attention network are used to process a stack of 2D image inputs to generate output scalars defining the location of a rectangular region of interest (ROI). The attention mechanism is used to make the model focus on the small number of informative slices in a stack. Then the smallest FOV that makes the neural network predicted ROI free of aliasing is calculated by an algebraic operation derived from MR sampling theory. We retrospectively collected 595 cases between February 2018 and February 2022. The framework's performance is examined quantitatively with intersection over union (IoU) and pixel error on position, and qualitatively with a reader study. We use the t-test for comparing quantitative results from all models and a radiologist. The proposed model achieves an average IoU of 0.867 and average ROI position error of 9.06 out of 512 pixels on 80 test cases, significantly better (P<0.05) than two baseline models and not significantly different from a radiologist (P>0.12). Finally, the FOV given by the proposed framework achieves an acceptance rate of 92% from an experienced radiologist.

Artifact- and content-specific quality assessment for MRI with image rulers

In clinical practice MR images are often first seen by radiologists long after the scan. If image quality is inadequate either patients have to return for an additional scan, or a suboptimal interpretation is rendered. An automatic image quality assessment (IQA) would enable real-time remediation. Existing IQA works for MRI give only a general quality score, agnostic to the cause of and solution to low-quality scans. Furthermore, radiologists' image quality requirements vary with the scan type and diagnostic task. Therefore, the same score may have different implications for different scans. We propose a framework with multi-task CNN model trained with calibrated labels and inferenced with image rulers. Labels calibrated by human inputs follow a well-defined and efficient labeling task. Image rulers address varying quality standards and provide a concrete way of interpreting raw scores from the CNN. The model supports assessments of two of the most common artifacts in MRI: noise and motion. It achieves accuracies of around 90%, 6% better than the best previous method examined, and 3% better than human experts on noise assessment. Our experiments show that label calibration, image rulers, and multi-task training improve the model's performance and generalizability.

Wasserstein GANs for MR Imaging: from Paired to Unpaired Training

Lack of ground-truth MR images (labels) impedes the common supervised training of deep networks for image reconstruction. To cope with this challenge, this paper leverages WGANs for unpaired training of reconstruction networks, where the inputs are the undersampled naively reconstructed images from one dataset, and the outputs are high-quality images from another dataset. The generator network is an unrolled neural network with a cascade of residual blocks and data consistency modules. The discriminator is also a multilayer CNN that plays the role of a critic scoring the quality of reconstructed images. Our extensive experiments with knee MRI datasets demonstrate unpaired WGAN training with minimal supervision is a viable option when there exists insufficient or no fully-sampled training label images that match the input images. Also, supervised paired training with additional WGAN loss achieves better and faster reconstruction compared to wavelet-based compressed sensing.