Style transfer for out-of-domain (OOD) speech synthesis aims to generate speech samples with unseen style (e.g., speaker identity, emotion, and prosody) derived from an acoustic reference, while facing the following challenges: 1) The highly dynamic style features in expressive voice are difficult to model and transfer; and 2) the TTS models should be robust enough to handle diverse OOD conditions that differ from the source data. This paper proposes GenerSpeech, a text-to-speech model towards high-fidelity zero-shot style transfer of OOD custom voice. GenerSpeech decomposes the speech variation into the style-agnostic and style-specific parts by introducing two components: 1) a multi-level style adaptor to efficiently model a large range of style conditions, including global speaker and emotion characteristics, and the local (utterance, phoneme, and word-level) fine-grained prosodic representations; and 2) a generalizable content adaptor with Mix-Style Layer Normalization to eliminate style information in the linguistic content representation and thus improve model generalization. Our evaluations on zero-shot style transfer demonstrate that GenerSpeech surpasses the state-of-the-art models in terms of audio quality and style similarity. The extension studies to adaptive style transfer further show that GenerSpeech performs robustly in the few-shot data setting.

Recently, text-to-speech (TTS) models such as FastSpeech and ParaNet have been proposed to generate mel-spectrograms from text in parallel. Despite the advantage, the parallel TTS models cannot be trained without guidance from autoregressive TTS models as their external aligners. In this work, we propose Glow-TTS, a flow-based generative model for parallel TTS that does not require any external aligner. By combining the properties of flows and dynamic programming, the proposed model searches for the most probable monotonic alignment between text and the latent representation of speech on its own. We demonstrate that enforcing hard monotonic alignments enables robust TTS, which generalizes to long utterances, and employing generative flows enables fast, diverse, and controllable speech synthesis.

Neural Transducer (e.g., RNN-T) has been widely used in automatic speech recognition (ASR) due to its capabilities of efficiently modeling monotonic alignments between input and output sequences and naturally supporting streaming inputs. Considering that monotonic alignments are also critical to text to speech (TTS) synthesis and streaming TTS is also an important application scenario, in this work, we explore the possibility of applying Transducer to TTS and more. However, it is challenging because it is difficult to trade off the emission (continuous mel-spectrogram prediction) probability and transition (ASR Transducer predicts blank token to indicate transition to next input) probability when calculating the output probability lattice in Transducer, and it is not easy to learn the alignments between text and speech through the output probability lattice. We propose SpeechTransducer (Speech-T for short), a Transformer based Transducer model that 1) uses a new forward algorithm to separate the transition prediction from the continuous mel-spectrogram prediction when calculating the output probability lattice, and uses a diagonal constraint in the probability lattice to help the alignment learning; 2) supports both full-sentence or streaming TTS by adjusting the look-ahead context; and 3) further supports both TTS and ASR together for the first time, which enjoys several advantages including fewer parameters as well as streaming synthesis and recognition in a single model. Experiments on LJSpeech datasets demonstrate that Speech-T 1) is more robust than the attention based autoregressive TTS model due to its inherent monotonic alignments between text and speech; 2) naturally supports streaming TTS with good voice quality; and 3) enjoys the benefit of joint modeling TTS and ASR in a single network.

We introduce Meta-Album, an image classification meta-dataset designed to facilitate few-shot learning, transfer learning, meta-learning, among other tasks. It includes 40 open datasets, each having at least 20 classes with 40 examples per class, with verified licences. They stem from diverse domains, such as ecology (fauna and flora), manufacturing (textures, vehicles), human actions, and optical character recognition, featuring various image scales (microscopic, human scales, remote sensing). All datasets are preprocessed, annotated, and formatted uniformly, and come in 3 versions (Micro \subset Mini \subset Extended) to match users' computational resources. The other 10 will be released shortly after.

This paper presents a solid piece of work on the speaker-dependent neural TTS system, building on previous works of Deep Voice and Tacotron architecture. The key idea is to learn a speaker-dependent embedding vector jointly with the neural TTS model. The paper is clearly written, and the experiments are presented well. My comments are as follows. ASR researchers later find that using fixed speaker embeddings such i-vectors can work equally well (or even better).

In this paper, we create benchmarks and assess the effectiveness of error correction methods for Japanese vouchers in OCR (Optical Character Recognition) systems. It is essential for automation processing to correctly recognize scanned voucher text, such as the company name on invoices. However, perfect recognition is complex due to the noise, such as stamps. Therefore, it is crucial to correctly rectify erroneous OCR results. However, no publicly available OCR error correction benchmarks for Japanese exist, and methods have not been adequately researched. In this study, we measured text recognition accuracy by existing services on Japanese vouchers and developed a post-OCR correction benchmark. Then, we proposed simple baselines for error correction using language models and verified whether the proposed method could effectively correct these errors. In the experiments, the proposed error correction algorithm significantly improved overall recognition accuracy.

Microsoft has released a white paper of sorts outlining what the company is doing to secure user data within Windows Recall, the controversial Windows feature that takes snapshots of your activity for later searching. As of late last night, Microsoft still hasn't said whether they will release Recall to the Windows Insider channels for further testing as originally planned. In fact, Microsoft's paper says very little about Recall as a product or when they will push Recall live to the public. Recall was first launched back in May as part of the Windows 11 24H2 update and it uses the local AI capabilities of Copilot PCs. The idea is that Recall captures periodic snapshots of your screen, then uses optical character recognition plus AI-driven techniques to translate and understand your activity.

Programming tutorials in the form of coding screencasts play a crucial role in programming education, serving both novices and experienced developers. However, the video format of these tutorials presents a challenge due to the difficulty of searching for and within videos. Addressing the absence of large-scale and diverse datasets for screencast analysis, we introduce the CodeSCAN dataset. It comprises 12,000 screenshots captured from the Visual Studio Code environment during development, featuring 24 programming languages, 25 fonts, and over 90 distinct themes, in addition to diverse layout changes and realistic user interactions. Moreover, we conduct detailed quantitative and qualitative evaluations to benchmark the performance of Integrated Development Environment (IDE) element detection, color-to-black-and-white conversion, and Optical Character Recognition (OCR). We hope that our contributions facilitate more research in coding screencast analysis, and we make the source code for creating the dataset and the benchmark publicly available at a-nau.github.io/codescan.

This paper presents a benchmark dataset for aligning lecture videos with corresponding slides and introduces a novel multimodal algorithm leveraging features from speech, text, and images. It achieves an average accuracy of 0.82 in comparison to SIFT (0.56) while being approximately 11 times faster. Using dynamic programming the algorithm tries to determine the optimal slide sequence. The results show that penalizing slide transitions increases accuracy. Features obtained via optical character recognition (OCR) contribute the most to a high matching accuracy, followed by image features. The findings highlight that audio transcripts alone provide valuable information for alignment and are beneficial if OCR data is lacking. Variations in matching accuracy across different lectures highlight the challenges associated with video quality and lecture style. The novel multimodal algorithm demonstrates robustness to some of these challenges, underscoring the potential of the approach.

In recent years, speech diffusion models have advanced rapidly. Alongside the widely used U-Net architecture, transformer-based models such as the Diffusion Transformer (DiT) have also gained attention. However, current DiT speech models treat Mel spectrograms as general images, which overlooks the specific acoustic properties of speech. To address these limitations, we propose a method called Directional Patch Interaction for Text-to-Speech (DPI-TTS), which builds on DiT and achieves fast training without compromising accuracy. Notably, DPI-TTS employs a low-to-high frequency, frame-by-frame progressive inference approach that aligns more closely with acoustic properties, enhancing the naturalness of the generated speech. Additionally, we introduce a fine-grained style temporal modeling method that further improves speaker style similarity. Experimental results demonstrate that our method increases the training speed by nearly 2 times and significantly outperforms the baseline models.