Qwen AI Glasses receive OTA update with call and media translation feature

The co-founder of Murf AI believes that the complexities associated with voice artificial intelligence in India have the potential to shape global technological trends. What Indian users consider ordinary conversation could become the most difficult test for voice AI, and solving this challenge will open up new possibilities worldwide.

Voice AI is capable of answering questions, confirming orders, or assisting with bookings. However, the question arises: can it correctly pronounce an Indian name, understand an address spoken in a mix of Hindi and English, and respond quickly enough to sound natural?

For Sneha Roy, co-founder and COO of Murf AI, such scenarios are not exceptional; they represent the baseline for creating voice AI that will function in India, and ultimately, globally.

During a masterclass at the TechSparks 2026 event titled 'Building Voice AI from India, for India and for the World,' Roy explained how India's linguistic complexity, infrastructure limitations, and economic considerations affect speech technology development.

She cited examples of products that were first adapted for Indian users before finding broader application. Google Maps implemented offline functionality for areas with limited internet access, and Netflix introduced a mobile-only plan that has since been adopted in 78 countries. Uber tested cash payments in Hyderabad before expanding this option to other markets.

Roy posits that voice AI can follow a similar path. For instance, when calling to check a delivery status in Bengaluru, a customer might mix English with a regional language, use a local pronunciation of an address, and expect the postal code to be spoken numerically.

A model built on standardized English may struggle with such nuances. Roy emphasized: 'If a customer has to change their manner of speaking so that the agent can cope, or constantly repeat their address or name, then we have not finished developing it.'

In her view, the task is to create systems that accommodate people's natural way of speaking, rather than requiring customers to adapt to the technology.

Natural speech is only one part of the problem. Voice agents must also react quickly and operate efficiently at scale when companies process millions of calls monthly. Roy reported that a complete voice agent pipeline, including speech recognition, language modeling, and speech generation, typically has a response budget of 500 to 800 milliseconds. She added that Murf has managed to reduce the latency of its text-to-speech model below 100 milliseconds, including network delays.

To achieve this, the company developed Falcon 2—a smaller speech model designed to run on a wider range of GPUs instead of depending on specific, high-performance hardware. Roy noted that this approach helped maintain low latency across different geographical regions while improving cost-effectiveness.

Similar engineering solutions could be beneficial for markets outside India, especially where computational infrastructure is limited or deploying voice agents at scale remains expensive.

Roy also highlighted the importance of a responsible approach to voice data. She mentioned that Murf collaborates with over 400 voice actors globally through long-term, royalty-based partnerships and does not clone voices without consent.

For clients with strict privacy requirements, the company offers APIs that can process data without storing it, as well as on-premise deployments for sensitive applications. Roy's central thesis is that India's linguistic diversity and infrastructural constraints should not be viewed as secondary requirements; they can help define what voice AI must do before it is ready for widespread adoption. As she put it: 'India is the capital of world specifications.'

Lawaken has launched the pre-sale of its X-AI Memory smart glasses. The basic model in a classic gunmetal wire frame is available from 699 yuan, while other frame options are priced at 899 yuan, and models with a titanium bridge cost 1,099 yuan.

Initially presented at WAIC 2026 in July as a constant personal memory device rather than a camera-focused gadget, the pre-sale starting in October maintains a frame weight of about 26 grams, an optional 10-gram 12-megapixel camera module, and extended recording focus time. Furthermore, the announcement specifies details on agent integration and pricing.

According to Lawaken's statements, the glasses can continuously record data for up to 12 hours on a single charge, which recharges in approximately 60 minutes. They also provide real-time translation into over 128 languages and transcribe 25 languages plus 21 Chinese dialects. The software is designed to extract tasks, key insights, and priorities from conversations, after which it can respond to voice queries using personal memory storage. Integration with Tencent Cloud WorkBuddy allows this storage to be converted into meeting protocols, plans, and drafts.

Thanks to the Model Context Protocol, Lawaken asserts that memory can be called from agent platforms such as Doubao, Feishu, Alibaba's QwenWork, Codex, Claude, and others. Users have the ability to edit, delete, or export saved memories, with cloud storage protected by AES-256 encryption.

Compared to the debut at WAIC in July, the pre-sale package emphasizes MCP hooks and document generation via WorkBuddy, and lists six frame options with retail prices starting from 699 yuan. A brief overview at WAIC noted Wake-AI 3.0 and the WakeeMemory capture-structuring-extraction-action cycle, a 109-degree camera field of view, 1080p video, and IP5 module protection; however, these camera specifications were not repeated in the October consumer announcements analyzed by Pandaily. The frame weight remains around 26 grams without lenses or a camera module, and quick-release frames for different styles are provided.

Lawaken positions these glasses for use in meetings, field inspections, and multi-threaded projects. As with other AI wearable devices, battery efficiency, transcription accuracy, and agent reliability will depend on real-world usage outside of laboratory claims.