Automatic song writing is a topic of significant practical interest. However, its research is largely hindered by the lack of training data due to copyright concerns and challenged by its creative nature. Most noticeably, prior works often fall short of modeling the cross-modal correlation between melody and lyrics due to limited parallel data, hence generating lyrics that are less singable. Existing works also lack effective mechanisms for content control, a much desired feature for democratizing song creation for people with limited music background. In this work, we propose to generate pleasantly listenable lyrics without training on melody-lyric aligned data. Instead, we design a hierarchical lyric generation framework that disentangles training (based purely on text) from inference (melody-guided text generation). At inference time, we leverage the crucial alignments between melody and lyrics and compile the given melody into constraints to guide the generation process. Evaluation results show that our model can generate high-quality lyrics that are more singable, intelligible, coherent, and in rhyme than strong baselines including those supervised on parallel data.
The emergence of generative large language models (LLMs) raises the question: what will be its impact on crowdsourcing. Traditionally, crowdsourcing has been used for acquiring solutions to a wide variety of human-intelligence tasks, including ones involving text generation, manipulation or evaluation. For some of these tasks, models like ChatGPT can potentially substitute human workers. In this study, we investigate, whether this is the case for the task of paraphrase generation for intent classification. We quasi-replicated the data collection methodology of an existing crowdsourcing study (similar scale, prompts and seed data) using ChatGPT. We show that ChatGPT-created paraphrases are more diverse and lead to more robust models.
While Transformer has become the de-facto standard for speech, modeling upon the fine-grained frame-level features remains an open challenge of capturing long-distance dependencies and distributing the attention weights. We propose \textit{Progressive Down-Sampling} (PDS) which gradually compresses the acoustic features into coarser-grained units containing more complete semantic information, like text-level representation. In addition, we develop a representation fusion method to alleviate information loss that occurs inevitably during high compression. In this way, we compress the acoustic features into 1/32 of the initial length while achieving better or comparable performances on the speech recognition task. And as a bonus, it yields inference speedups ranging from 1.20$\times$ to 1.47$\times$. By reducing the modeling burden, we also achieve competitive results when training on the more challenging speech translation task.
We propose DocFormerv2, a multi-modal transformer for Visual Document Understanding (VDU). The VDU domain entails understanding documents (beyond mere OCR predictions) e.g., extracting information from a form, VQA for documents and other tasks. VDU is challenging as it needs a model to make sense of multiple modalities (visual, language and spatial) to make a prediction. Our approach, termed DocFormerv2 is an encoder-decoder transformer which takes as input - vision, language and spatial features. DocFormerv2 is pre-trained with unsupervised tasks employed asymmetrically i.e., two novel document tasks on encoder and one on the auto-regressive decoder. The unsupervised tasks have been carefully designed to ensure that the pre-training encourages local-feature alignment between multiple modalities. DocFormerv2 when evaluated on nine datasets shows state-of-the-art performance over strong baselines e.g. TabFact (4.3%), InfoVQA (1.4%), FUNSD (1%). Furthermore, to show generalization capabilities, on three VQA tasks involving scene-text, Doc- Formerv2 outperforms previous comparably-sized models and even does better than much larger models (such as GIT2, PaLi and Flamingo) on some tasks. Extensive ablations show that due to its pre-training, DocFormerv2 understands multiple modalities better than prior-art in VDU.
Flow-based generative models are widely used in text-to-speech (TTS) systems to learn the distribution of audio features (e.g., Mel-spectrograms) given the input tokens and to sample from this distribution to generate diverse utterances. However, in the zero-shot multi-speaker TTS scenario, the generated utterances lack diversity and naturalness. In this paper, we propose to improve the diversity of utterances by explicitly learning the distribution of fundamental frequency sequences (pitch contours) of each speaker during training using a stochastic flow-based pitch predictor, then conditioning the model on generated pitch contours during inference. The experimental results demonstrate that the proposed method yields a significant improvement in the naturalness and diversity of speech generated by a Glow-TTS model that uses explicit stochastic pitch prediction, over a Glow-TTS baseline and an improved Glow-TTS model that uses a stochastic duration predictor.
Accurate syntactic representations are essential for robust generalization in natural language. Recent work has found that pre-training can teach language models to rely on hierarchical syntactic features - as opposed to incorrect linear features - when performing tasks after fine-tuning. We test what aspects of pre-training are important for endowing encoder-decoder Transformers with an inductive bias that favors hierarchical syntactic generalizations. We focus on architectural features (depth, width, and number of parameters), as well as the genre and size of the pre-training corpus, diagnosing inductive biases using two syntactic transformation tasks: question formation and passivization, both in English. We find that the number of parameters alone does not explain hierarchical generalization: model depth plays greater role than model width. We also find that pre-training on simpler language, such as child-directed speech, induces a hierarchical bias using an order-of-magnitude less data than pre-training on more typical datasets based on web text or Wikipedia; this suggests that in cognitively plausible language acquisition settings, neural language models may be more data-efficient than previously thought.
Recent years have witnessed considerable achievements in editing images with text instructions. When applying these editors to dynamic scene editing, the new-style scene tends to be temporally inconsistent due to the frame-by-frame nature of these 2D editors. To tackle this issue, we propose Control4D, a novel approach for high-fidelity and temporally consistent 4D portrait editing. Control4D is built upon an efficient 4D representation with a 2D diffusion-based editor. Instead of using direct supervisions from the editor, our method learns a 4D GAN from it and avoids the inconsistent supervision signals. Specifically, we employ a discriminator to learn the generation distribution based on the edited images and then update the generator with the discrimination signals. For more stable training, multi-level information is extracted from the edited images and used to facilitate the learning of the generator. Experimental results show that Control4D surpasses previous approaches and achieves more photo-realistic and consistent 4D editing performances. The link to our project website is https://control4darxiv.github.io.
Long-form question answering (LFQA) enables answering a wide range of questions, but its flexibility poses enormous challenges for evaluation. We perform the first targeted study of the evaluation of long-form answers, covering both human and automatic evaluation practices. We hire domain experts in seven areas to provide preference judgments over pairs of answers, along with free-form justifications for their choices. We present a careful analysis of experts' evaluation, which focuses on new aspects such as the comprehensiveness of the answer. Next, we examine automatic text generation metrics, finding that no existing metrics are predictive of human preference judgments. However, some metrics correlate with fine-grained aspects of answers (e.g., coherence). We encourage future work to move away from a single "overall score" of the answer and adopt a multi-faceted evaluation, targeting aspects such as factuality and completeness. We publicly release all of our annotations and code to spur future work into LFQA evaluation.
We introduce Three Towers (3T), a flexible method to improve the contrastive learning of vision-language models by incorporating pretrained image classifiers. While contrastive models are usually trained from scratch, LiT (Zhai et al., 2022) has recently shown performance gains from using pretrained classifier embeddings. However, LiT directly replaces the image tower with the frozen embeddings, excluding any potential benefits of contrastively training the image tower. With 3T, we propose a more flexible strategy that allows the image tower to benefit from both pretrained embeddings and contrastive training. To achieve this, we introduce a third tower that contains the frozen pretrained embeddings, and we encourage alignment between this third tower and the main image-text towers. Empirically, 3T consistently improves over LiT and the CLIP-style from-scratch baseline for retrieval tasks. For classification, 3T reliably improves over the from-scratch baseline, and while it underperforms relative to LiT for JFT-pretrained models, it outperforms LiT for ImageNet-21k and Places365 pretraining.
There are significant challenges for speaker adaptation in text-to-speech for languages that are not widely spoken or for speakers with accents or dialects that are not well-represented in the training data. To address this issue, we propose the use of the "mixture of adapters" method. This approach involves adding multiple adapters within a backbone-model layer to learn the unique characteristics of different speakers. Our approach outperforms the baseline, with a noticeable improvement of 5% observed in speaker preference tests when using only one minute of data for each new speaker. Moreover, following the adapter paradigm, we fine-tune only the adapter parameters (11% of the total model parameters). This is a significant achievement in parameter-efficient speaker adaptation, and one of the first models of its kind. Overall, our proposed approach offers a promising solution to the speech synthesis techniques, particularly for adapting to speakers from diverse backgrounds.