Multicultural societies involve complex daily conversations. In vibrant, multicultural hubs like Hong Kong, daily conversations wave effortlessly between Cantonese, English and Mandarin. A business meeting might shift languages in a heartbeat. Similarly, in Malaysia, English often dominates commerce, but side conversations might flow in Bahasa Malaysia, Mandarin or Tamil. This dynamic code-switching, a hallmark of diverse metropolises, fosters connection and accommodates varied linguistic backgrounds, shaped by immigration, regional influences and global pop culture. Yet, it can also create confusion, as non-professional interpreters struggle to convey nuances, risking miscommunications and lost trust.  

Fano, a provider of AI-based speech recognition services, aims to address this issue. Despite operating in a competitive space, Fano distinguishes itself through its extensive research, which underpins its solutions, its ability to handle the granularity of mixed language scenarios, and its capacity for insight generation based on the content it processes. Its solutions are domain-specific catering to the clients’ industries, with a focus on helping organisations understand customer intent, identify opportunity and guard against risk.

Tackling low-resource languages and diverse accents

Based in Hong Kong, the 10-year-old company is the product of real-world challenges that its founders — former researchers and professors at Hong Kong University — saw around them every day. It bills its solutions as globally applicable but Asia-focused, with support for 15-20 languages and emphasis on “low resource” ones like Cantonese. Despite being spoken by approximately 85 million people across Hong Kong, Macau, China’s Canton region, and overseas Chinese nationals, transcription data for Cantonese remains limited. 

This scarcity is compounded by the wide range of accents and intonations found in multicultural hubs like Hong Kong. As an international financial centre and a premier innovation and technology centre, the city attracts professionals from around the world. While their accents vary, many embrace the local language as part of their immersion in the community. Fano’s strategy involves gathering extensive data from all regions where the language is spoken to enable a robust model that recognises a wide range of intonations.

From Hong Kong roots to compliance-driven growth

Fano claims to process mixed language interactions that involve frequent switching with over 90% accuracy and can cut through and decipher thickly accented English in a way that others may not be able to.

“The language environment in Hong Kong is very complicated — too intricate for traditional AI to handle,” explains Dr Albert Lam, Fano’s Chief Technology Officer and Chief Scientist. “We can’t control how people speak, so we saw the need for technology to address these mixed language scenarios, not just identify what was being said, but truly understand the details.”

Having completed a Series B funding round in May 2024, Fano is in expansion mode, with its eyes on Southeast Asia and the Middle East. In every market it operates in, finance and the public sector are core segments. Within those, as well as customer service applications, supporting compliance and risk management has become an important area. Its emergence was partly driven by the pandemic. Unable to pursue international growth plans under lockdown, Fano started to explore new ways to provide assistance to organisations in Hong Kong. 

“We thought, why don’t we explore how conversations evolve in banks? That led us to step into compliance, especially sales compliance,” says Dr Lam. One example of this is analysing conversations for omissions such as disclaimers and risk disclosures to avoid penalties from regulators. As non-banking financial organisations come under greater scrutiny, the breadth of companies it helps stay compliant looks likely to widen.

Fighting fraud and deepfakes with voice technology

Another potential application involves detecting fraudulent intent through voice screening of natural conversational speech using deep neural networks (DNN) as opposed to a pre-determined phrase. The process involves verifying someone’s identity using voice biometrics technology and it is content-independent, meaning the verification process can be completed while the conversation goes on. There are multiple benefits to this versatile approach, including the prevention of so-called replay attacks, where a bad actor gains access to a system by using recorded speech.

The need for technology to recognise the fine details of an individual’s voice is becoming pronounced as AI becomes more sophisticated in its ability to mimic the voices of others. Dr Lam says Fano has developed an “audio-defect model” to distinguish AI-generated voice content. He notes that this kind of defence technology exists in a constant state of evolution akin to antivirus software given the pace at which AI, and with it, deepfake capabilities, are advancing.

Moving towards agentic AI in customer service

Fano’s customer service support is evolving too. “We are moving to the agentic world,” Dr Lam says. Where AI is currently used to reduce the demands on human agents by performing relatively simple tasks, Fano envisages “AI agents” with far greater capacity for logic, decision making and overall autonomy. Key to this is the ability to reason — the interstitial between input in the form of data from a multitude of past conversations, and the ultimate output. 

Dr Lam sees this as the basis for “a group of AI agents working together” with different functions, some high-level and some lower, much like a human team. He underscores that while this would potentially save manpower and reduce costs, the focus would be on “helping customers achieve breakthroughs” rather than downsize their staff.

Fano’s ambitious future is closely entwined with Hong Kong Science and Technology Parks Corporation (HKSTP). The organisation has been instrumental in helping Fano engage with customers in sectors and markets outside its immediate sphere, including those in the public service sector. Fano is currently working with HKSTP to explore the potential of its solutions for healthcare, as well as raising its profile in markets such as the UAE, Singapore, the UK and Europe.

As technology’s role in society expands, human and linguistic complexity remain a fact of life. Fano both celebrates this reality and offers practical tools to navigate it.

—

Want updates like this delivered directly? Join our WhatsApp channel and stay in the loop.

e27 and HKSTP collaborated on this article.

We can share your story at e27 too! Engage the Southeast Asian tech ecosystem by bringing your story to the world. You can reach out to us here to get started.

Featured Image Credit: HKSTP

The post Cracking the code-switch: How a Hong Kong AI firm helps turn linguistic chaos into commercial clarity appeared first on e27.