Voice AI Weekly #10 - Third Week of September 2026
Over the past week, infrastructure for regulated industries and changes in the voice policies of major platforms stood out in the overseas Voice AI industry. Here is a summary of five news updates.
1. India's Navana.ai Secures KRW 4 Billion in Series A for BFSI-Specialized Voice AI
Indian voice AI startup Navana.ai raised 40 crore rupees (approx. USD 4.2 million) in a Series A funding round led by Ronnie Screwvala on September 15. Based on its proprietary voice model supporting 12 Indian languages and 45 dialects, the company has processed over 100 million minutes of voice AI and secured clients in the financial sector, including Bajaj Finserv and Ujjivan Small Finance Bank. The investment will be used to strengthen its AI contact center platform and expand on-premises deployment.
In markets with highly segmented languages and dialects, specialized voice AI that prioritizes data protection and regulatory compliance is gaining more attention than general-purpose models. This case demonstrates that a strategy combining Korean specialization with on-premises deployment could also show potential in overseas markets.
Navana.ai Raises ₹40 Crore in Series A to Scale Voice AI for BFSI
2. OpenAI Releases GPT-Live-1 as an API
On September 10, OpenAI released GPT-Live-1, the full-duplex model that powers ChatGPT's voice mode, as an API. Developers can connect their preferred inference models and tool-calling harnesses to this model, which integrates listening and speaking. It also provides capabilities to adjust response speed and expression by reflecting the speaker's tone and emotion. Key features such as background noise cancellation and interruption handling were also highlighted.
3. Qualcomm Unveils New NPU Targeting On-Device Voice AI
Ahead of the Snapdragon Summit on September 22, Qualcomm pre-released the Hexagon NPU specifications of its next-generation flagship chip, the Snapdragon 8 Elite Gen 6 Pro, on September 11. It features the newly embedded Element Accelerator, a dedicated accelerator for Transformer models, and has increased shared memory by 50%. As a result, MoE models with up to 30 billion parameters can run directly on the device.
The hardware capacity to handle voice recognition and even sentiment analysis on-device is expanding. The demand for implementing on-device voice AI in closed networks or security-critical environments is highly likely to grow even further in the future.
Qualcomm's next flagship chip is getting a serious AI brain upgrade
4. US Department of Veterans Affairs Fully Expands Ambient Voice Scribes
On September 14, the US Department of Veterans Affairs (VA) announced that it is expanding its ambient voice scribe pilot program—which began in October last year across 10 medical centers—to primary care, mental health, rehabilitation medicine, and all specialty departments. Scribes from Abridge and Knowtex record medical conversations upon patient consent and automatically draft Electronic Health Records (EHRs); they have been used in over 980,000 clinical encounters to date. The VA plans to sign an enterprise-wide contract within this year to expand this across outpatient care.
While maintaining the procedure where doctors review and approve records, they successfully applied consent-based recording and domain-specific summarization to large-scale medical settings. This case shows that designing accuracy and verification procedures together is essential to lead to actual adoption and approval.
VA to gradually expand medical AI scribe following successful pilots
5. Google Assistant Departs, Gemini Takes Over
Starting September 4, Google began phasing out Google Assistant on Android, Wear OS, earbuds, and Android Auto, transitioning users to Gemini. The transition is taking place over several weeks, and once users switch to Gemini, they cannot revert to the previous system.
A significant portion of Android users worldwide will now use an LLM-based voice assistant as their default. As consumer voice assistants become the standard, companies will inevitably have to think deeper about how to differentiate their own service's voice experience.
Google Assistant to be replaced by Gemini starting September on Android and WearOS
It was a week where platform voice assistants underwent restructuring, and in regulated industries, domain specialization and on-premises solutions successfully drove both investment and adoption approvals.
If you want to know more details or are curious about MAGO's technology, please contact us at contact@holamago.com. We will be back next week with more interesting industry news!