Shashwat Verma

Alexander H. Liu, Alexis Tacnet, Andy Ehrenberg et al. · deepmind, tsinghua

We introduce Voxtral TTS, an expressive multilingual text-to-speech model that generates natural speech from as little as 3 seconds of reference audio. Voxtral TTS adopts a hybrid architecture that combines auto-regressive generation of semantic speech tokens with flow-matching for acoustic tokens. These tokens are encoded and decoded with Voxtral Codec, a speech tokenizer trained from scratch with a hybrid VQ-FSQ quantization scheme. In human evaluations conducted by native speakers, Voxtral TTS is preferred for multilingual voice cloning due to its naturalness and expressivity, achieving a 68.4\% win rate over ElevenLabs Flash v2.5. We release the model weights under a CC BY-NC license.

Liu He, Xiao Zeng, Yizhi Song et al.

Multimodal Large Language Models (MLLMs) struggle with accurately capturing camera-object relations, especially for object orientation, camera viewpoint, and camera shots. This stems from the fact that existing MLLMs are trained on images with limited diverse camera-object relations and corresponding textual descriptions. To address this, we propose a synthetic generation pipeline to create large-scale 3D visual instruction datasets. Our framework takes 3D assets as input and uses rendering and diffusion-based image generation models to create photorealistic images preserving precise camera-object relations. Additionally, large language models (LLMs) are used to generate text prompts for guiding visual instruction tuning and controlling image generation. We create Ultimate3D, a dataset of 240K VQAs with precise camera-object annotations, and corresponding benchmark. MLLMs fine-tuned on our proposed dataset outperform commercial models by a large margin, achieving an average accuracy improvement of 33.4% on camera-object relation recognition tasks. Our code, dataset, and benchmark will contribute to broad MLLM applications.

Manpreet Kaur, Ankur Tomar, Srijan Mishra et al.

Deep Learning methods are highly local and sensitive to the domain of data they are trained with. Even a slight deviation from the domain distribution affects prediction accuracy of deep networks significantly. In this work, we have investigated a set of techniques aimed at increasing accuracy of generator networks which perform translation from one domain to the other in an adversarial setting. In particular, we experimented with activations, the encoder-decoder network architectures, and introduced a Loss called cyclic loss to constrain the Generator network so that it learns effective source-target translation. This machine learning problem is motivated by myriad applications that can be derived from domain adaptation networks like generating labeled data from synthetic inputs in an unsupervised fashion, and using these translation network in conjunction with the original domain network to generalize deep learning networks across domains.