The Fire and Disaster Management Agency will introduce an artificial intelligence-powered system to automatically transcribe emergency calls with the aim of streamlining record-keeping and easing the workload on staff amid rising ambulance dispatches. The Fire and Disaster Management Agency plans to launch a model project in fiscal 2027 to utilize artificial intelligence in the handling of 119 emergency calls for firefighting, rescue and ambulance services. The agency will introduce an AI-powered speech recognition system that automatically transcribes emergency calls with the aim of streamlining the process of recording call details and easing the burden on staff amid a growing number of ambulance dispatches. Related expenses will be included in the agency's fiscal 2027 budget request. The project will utilize so-called "vertical AI," which specializes in specific fields. The AI system will also be equipped with a function that advises staff on necessary actions according to the content of each call.

Google is making it easier for people to support their favorite publications by introducing a new "Preferred Sources" button webmasters can add to their websites. "When readers click this button, it adds the site as a Preferred Source on Google and immediately directs them back to where they left off on the publisher's page," Google explains. After clicking the button, you'll see the publication appear more frequently Google Search, including in Top Stories, AI Overviews and AI Mode. The button formalizes something publishers, including Engadget, had been independently adding their websites for a while now. Even with referral traffic from Google declining in recent years due to products like AI Overviews and AI Mode, the company's search engine remains an important source of revenue for many media outlets.

When you purchase through links in our articles, we may earn a small commission. While you can set the Gemini app to create images and videos without a visible watermark, hidden watermarks and metadata will remain. On a week when AI watermarking is a hot topic, Google has just announced that Gemini will let you drop the visible watermarks from created images and videos. Of course, "visible" is the key word here, with Google exec Josh Woodward explaining in a social post that invisible SynthID watermarks and metadata will remain embedded in created Gemini images. "We're striking a balance here between creative control and safety," Woodard wrote on X, adding that the option to drop the visible watermark is available for the Nano Banana image model, the Omni model for videos, and Lyria for generated music.

Google's Pixel Camera chief, Isaac Reynolds, thinks year-to-year upgrades to smartphone cameras are still critical, even though some detractors think they're passé. "A person who is a casual photographer and desperately wants to be a hobbyist--they're never going to break that ceiling unless you give them the tools, he says. The features that others call'meh'--for some people, those features are life-changing. Take the 30X Super Res Zoom on the Pixel 7 Pro from 2022, which uses machine learning to stitch together multiple frames in a digitally zoomed photo to give you a result that's much sharper than you're used to. Reynolds says that because of this feature, Google now sees people using 30X zoom dramatically more than before. The new Pixel 11 Pro, announced today, can zoom up to 120X, though in addition to the digital stitching, it relies on AI-generated pixels to fill in some details. Reynolds says maybe having that capability is enough to get someone interested in, say, wildlife photography. After experimenting on their phone, they'd be ready to invest in a professional camera and a big zoom lens. "There's a reason I work on low light and zoom every single year, he says.

During Sonos' third quarter financial call, the company said that it was planning a product launch in September. Based on a new FCC filing spotted by Lowpass, one of the products it's unveiling at the event could be a pair of headphones called the "Ace Ultra." While the filing is heavily redacted, the Ace Ultra seems to have an over-ear design, similar to the original Sonos Ace launched in 2024. But it will be available in five colors instead of just two, namely Greige (or light gray), Shadow Black, Agave, Blush and Sand. The FCC filing also hinted that the device will support Sonos Voice Control, the company's voice assistant.

A few simple adjustments to your settings will opt you out. Google is at it again . The company recently, and quietly, introduced a change to how it hoovers up our data to train its AI platforms. It can now scoop up media you upload to its various search tools for training purposes, according to a report by . This includes images, files and audio and video recordings.

We introduce SLED, an alternative approach to speech language modeling by encoding speech waveforms into sequences of continuous latent representations and modeling them autoregressively using an energy distance objective. The energy distance offers an analytical measure of the distributional gap by contrasting simulated and target samples, enabling efficient training to capture the underlying continuous autoregressive distribution. By bypassing reliance on residual vector quantization, SLED avoids discretization errors and eliminates the need for the complicated hierarchical architectures common in existing speech language models.

We propose Shallow Flow Matching (SFM), a novel mechanism that enhances flow matching (FM)-based text-to-speech (TTS) models within a coarse-to-fine generation paradigm. Unlike conventional FM modules, which use the coarse representations from the weak generator as conditions, SFM constructs intermediate states along the FM paths from these representations. During training, we introduce an orthogonal projection method to adaptively determine the temporal position of these states, and apply a principled construction strategy based on a singlesegment piecewise flow. The SFM inference starts from the intermediate state rather than pure noise, thereby focusing computation on the latter stages of the FM paths. We integrate SFM into multiple TTS models with a lightweight SFM head. Experiments demonstrate that SFM yields consistent gains in speech naturalness across both objective and subjective evaluations, and significantly accelerates inference when using adaptive-step ODE solvers. Demo and codes are available at https://ydqmkkx.github.io/SFMDemo/.

Attention-based encoder decoder models remain a popular choice for state-of-the-art automatic speech recognition (ASR). These models combine a powerful audio encoder that extracts rich acoustic features with a decoder that autoregressively produces the ASR output. The decoder handles two critical tasks: (1) building rich text-only context and (2) merging acoustic information from the encoder to ensure the predictions remain faithful to the audio. We observe a systematic pattern across the attention distributions of decoder layers in prior architectures: the initial layers direct most attention towards building textual context, while the later layers largely focus on merging acoustic and textual information for the final predictions. Leveraging this key insight, we propose BLOCKDECODER, a novel decoder architecture comprising two distinct components: a text encoder that is purely text-based, and a MERGER that combines information from the audio encoder and text encoder to generate output tokens. Unlike traditional decoders, the MERGER autoregressively predicts a sequence of K tokens within a block of size K, while relying on the same precomputed contextual information from both text and audio encoders across the block. This design choice allows for the efficient reuse of encoder representations. The separation of the decoder into the text encoder and the MERGER promotes modularity and more flexible control of parameters via the number of text encoder and MERGER layers. As a result, BLOCKDECODER yields a significant speedup ( 2x) compared to traditional decoders, across diverse datasets, languages, and speech tasks, without any degradation in performance.

Physiological signals are often corrupted by motion artifacts, baseline drift, and other low-SNR disturbances, which pose significant challenges for analysis. Additionally, these signals exhibit strong non-stationarity, with sharp peaks and abrupt changes that evolve continuously, making them difficult to represent using traditional time-domain or filtering methods. To address these issues, a novel waveletbased approach for physiological signal analysis is presented, aiming to capture multi-scale time-frequency features in various physiological signals. Leveraging this technique, two large-scale pretrained models specific to EMG and ECG are introduced for the first time, achieving superior performance and setting new baselines in downstream tasks. Additionally, a unified multi-modal framework is constructed by integrating pretrained EEG model, where each modality is guided through its dedicated branch and fused via learnable weighted fusion. This design effectively addresses challenges such as low signal-to-noise ratio, high inter-subject variability, and device mismatch, outperforming existing methods on multi-modal tasks. The proposed wavelet-based architecture lays a solid foundation for analysis of diverse physiological signals, while the multi-modal design points to nextgeneration physiological signal processing with potential impact on wearable health monitoring, clinical diagnostics, and broader biomedical applications.