Voice AI in vogue: Nvidia releases new open transcription model

techcrunch.com Oct 29, 2026

Voice AI in vogue: Nvidia releases new open transcription model

Curated by Shubham · TechCapsules · Oct 29, 2026

AI-summarised brief · reviewed before publication

Nvidia unveiled Nemotron 3.5 Transcribe, a free, open‑weight model that transcribes audio in up to 40 languages and translates 31 languages to and from English. The 30‑billion‑parameter model handles recordings up to an hour long, distinguishes up to eight speakers, and claims top performance on the Hugging Face Open ASR Leaderboard. Nvidia also released Granary v2, an expanded open‑source speech dataset, and promotes integration with its NeMo Speech toolkit for industry‑specific jargon. The model runs on a single chip, from Jetson AGX Thor to Blackwell data‑center GPUs.

💡 Why It Matters

· By offering a high‑performance, multilingual transcription tool that scales from edge devices to data‑center GPUs, Nvidia expands the practical reach of AI‑powered voice applications in business and healthcare, sectors that demand reliable, industry‑specific speech recognition.

AI Capsules Nvidia

Read full article on techcrunch.com

AI execs may be too out of touch to build for the…

Former Yahoo CEO Marissa Mayer has drawn criticism for her AI assistant, Dazzle, after revealing it generated a generic description of her ten‑year‑old daughter. Mayer showcased the app’s ability to pull photos and suggest birthday themes, sparking debate over AI’s role in child‑rearing and personal relationships. The backlash echoes broader concerns about Silicon Valley’s perceived detachment from everyday life, as leaders like Sam Altman and Mark Zuckerberg promote AI‑driven social solutions. The controversy highlights tensions between technological ambition and societal values.

💡 The episode underscores a growing mistrust of AI in intimate, human‑centered contexts, challenging the narrative that automation can seamlessly replace emotional labor. It forces a reckoning on how tech leaders shape public expectations about the future of family and connection.

9to5mac.com · Oct 28

Perplexity upgrades Mac app with ChatGPT-powered voice mode

Perplexity has integrated OpenAI’s gpt‑live‑1 voice model into its Mac app, enabling the assistant to listen while it speaks. The update, released today, adds multitasking capabilities, allowing users to manage multiple computer sessions, start new tasks, and create projects within a single voice session. The feature is available to all Pro and Max subscribers on both the Mac app and web. Perplexity also recently introduced an offline AI mode that runs locally on Macs, enhancing privacy and offline functionality.

💡 By embedding a bidirectional voice model, Perplexity turns its desktop assistant into a more natural, hands‑free tool, potentially reshaping how professionals interact with AI on their primary workstation.

ign.com · Oct 28

Masahiro Sakurai Shrugs Off AI Games as ‘Unprofitable Titles’ as Clones and…

Masahiro Sakurai, creator of Super Smash Bros, told Denfaminico Gamer that the surge of AI‑generated games will mostly yield “unprofitable titles” lacking the originality of human‑crafted works. He noted Steam’s catalog grew from 275 releases in 2010 to about 19,000 today, and predicts AI will push the number even higher. Sakurai argued AI can only remix existing mechanics, citing simple puzzle concepts such as “Needle in a Haystack Simulator” as examples of easy replication. He contrasted this with the distinctive artistic vision of developers like Fumito Ueda, whose future projects he believes AI cannot emulate. Sakurai concluded that creators who forge unique styles will become increasingly vital as the market floods with low‑quality, mass‑produced games for players seeking depth today.

💡 As AI churns out endless copycat titles, the market will reward the few creators who can deliver truly novel experiences, reshaping where investment and player attention flow.

semiconductor-digest.com · Oct 28

Why Is AI Finally Sticking Inside the Fab?

Semiconductor manufacturers are increasingly embedding artificial‑intelligence tools in specific process steps rather than deploying a single fab‑wide platform, analyst Mahima Sambre reports. Companies fund narrow, deterministic applications such as defect classifiers on inspection equipment and virtual metrology models for individual deposition stages, targeting the most costly yield losses. Acumen Research estimates the AI‑in‑manufacturing market at $6.5 billion in 2025, rising to $26.8 billion by 2035, a 15.2 % compound annual growth rate that outpaces related hardware. Asia‑Pacific accounts for 55 % of inspection‑system sales, while AI is also used to preserve senior engineers’ tacit knowledge as they retire. Emerging concerns include export‑control limits on data processing, the electricity demand of on‑tool inference, and recent acquisitions of metrology‑software teams. Sambre concludes that autonomous fabs remain distant.

💡 Capturing retiring engineers’ expertise with AI offers a rare, time‑critical knowledge transfer that could sustain yield improvements as the industry scales.

thehackernews.com · Oct 27

AI-Generated Fake Login Pages Are Making Account Takeovers Easier

Generative AI enables attackers to craft highly realistic fake login pages that mimic trusted brands, capturing usernames, passwords, MFA codes, and even session tokens. These AI‑generated replicas use company logos, layouts, and personalized phishing messages to deceive employees, making detection harder even with MFA. Verizon’s 2026 Data Breach Investigations Report shows phishing accounts for 44% of AI‑assisted initial‑access attacks. Traditional email filters and URL blocklists often miss new look‑alike domains, underscoring the need for advanced web‑site evaluation tools.

💡 The rise of AI‑powered phishing threatens to bypass multi‑factor authentication, exposing corporate accounts to rapid, large‑scale breaches that traditional defenses cannot detect.

semiconductor-digest.com · Oct 27

Has AI Shifted the Sustainability Balance from Chip Design to Manufacturing?

The semiconductor industry’s sustainability focus is shifting from chip design to manufacturing as AI demand drives new fabs. Each new node consumes more water, energy, materials, and chemicals, with imec warning that IC production could reach 3% of global greenhouse‑gas emissions by 2040. Blyler cites industry partners—IDA Ireland, JST, ZEISS SMT, TechInsights—to outline solutions: co‑optimizing design and fab processes, adopting renewable energy, improving water recycling, using lower‑emission gases, and extending equipment life to reduce waste. TechInsights forecasts that between 2028 and 2030, AI accelerator shipments will rise only 2.8% while manufacturing emissions could increase 40.4%, reflecting a shift toward carbon‑intensive architectures. A proposed feedback loop leverages AI for process control, predictive maintenance, virtual fabs, and digital twins to curb the environmental footprint of AI chip production.

💡 The move places the environmental burden on fabs, potentially making semiconductor manufacturing a major climate contributor. Addressing this shift is essential for meeting global emissions targets and sustaining AI growth.