Hello, this is MAGO. Every week, we gather and deliver the latest Voice Tech news happening all around the world over the past week.

We have selected 5 international voice technology news stories for this week that you would not want to miss.

1. Nvidia Bets on Paris Voice AI Startup Gradium - $100 Million Seed

Gradium, founded by researchers from the French open-source voice AI lab Kyutai, has secured a seed round of over $100 million with Nvidia's participation. This is the result just 7 months after its founding. Gradium is building a low-latency voice model infrastructure.

It appears that massive capital, which had been heavily focused on LLMs, is now moving down to the voice layer. This seems to indicate that the industry has started to view voice technology not just as a simple service (application), but as a core 'infrastructure'.

Source: TechCrunch · PYMNTS

2. Apple On-Device STT API 'SpeechAnalyzer' Surpasses Whisper Small

Apple's on-device speech recognition API 'SpeechAnalyzer', introduced for iOS 26, outperformed OpenAI's Whisper Small in benchmarks. Based on LibriSpeech, the Word Error Rate (WER) is 2.12% in clean audio and 4.56% in noisy audio, which is more accurate and 3 times faster than Whisper Small (3.74% / 7.95%). A major advantage is that it is processed entirely within the device without going through a server. However, among field developers, some comment that recognition of proper nouns and diverse accents is still lacking.

On-device STT has reached a level comparable to cloud-based models. The remaining gap lies in handling terminology and domain adaptation, which is an area that needs to be solved through domain fine-tuning rather than general-purpose models.

Source: GIGAZINE · Silicon Report

3. Taco Bell Expands Drive-Thru Voice AI to 900 Stores

Taco Bell is fully introducing 'Omilia's' voice ordering system to approximately 890 stores in the US. Although there was a lot of talk about noise recognition issues and ordering errors during the pilot test in 2024, they ultimately addressed these issues and moved to a large-scale commercialization phase.

With car engine sounds, short and fragmented sentences, and all kinds of menu names, the drive-thru is where all the worst conditions for voice recognition technology meet. The fact that a scale-up to 900 stores has been achieved here seems to be a signal that voice technology capable of withstanding real-world noise has now moved out of the lab and proven its marketability!

Source: Nation's Restaurant News · Fox Business

4. Kardome Voice AI Integrated into LG OLED TVs, Popularizing On-Device

The spatial voice processing technology from Kardome, a company specializing in voice technology, has entered people's living rooms by being integrated into LG OLED TVs. This is an example of applying a technology that accurately separates and recognizes only the user's voice, even in environments like a living room where multiple people speak simultaneously or TV noise is mixed in, to everyday home appliances.

The main stage for voice technology is rapidly expanding beyond smartphones to TVs, home appliances, and robots. From the perspective of device manufacturers, 'on-device voice engines' that eliminate concerns about cloud costs or data security seem to be transitioning from an option to a necessity.

Source: TMX Newsfile

5. "AI Voice Scams: Now a Compliance (Legal Regulation) Issue Beyond Security"

With the dawn of the era where AI agents can make phone calls at scale on behalf of humans, 'Reality Defender', a synthetic voice detection company, has warned of the dangers of 'Agentic Callers'. Indeed, in the United States, billions of dollars in losses have been reported due to Medicare scams exploiting fake voices. Consequently, a flood of analyses addressing corporate voice security measures is being released.

The moment the reliability of voice channels drops, the business operations of call centers, finance, and healthcare industries are directly hit. Moving beyond simply scanning text, the demand for technology that determines real versus fake at the voice signal layer is bound to explode.

Source: Biometric Update · New York Post

<This Week's Trend, One-Line Summary>

"Money is moving into voice infrastructure, deployment is going inside devices (on-device), and risks are shifting toward the credibility of voice channels."

If you have any further questions about the news covered today or are considering introducing voice technology into your business, feel free to email us anytime at contact@holamago.com.

MAGO will return next week with even richer news. Have a wonderful week, everyone!