Proto has released groundbreaking voice AI technology designed specifically for underserved local languages. This initiative, supported by partnerships with the Canadian government and Gates Foundation, represents a major step forward in bridging communication gaps in low and middle-income countries.
Building on R&D funding from Canada’s ScaleAI program, Proto is releasing voice recognition enhancements to its AI Customer Experience (AICX) platform with automated speech recognition (ASR) and text-to-speech (TTS) capabilities for languages such as Tagalog, Kinyarwanda, Cebuano, and Oshiwambo. These enhancements are aimed at improving accessibility and customer experience across government, healthcare, finance, and transport sectors.
This focus on voice AI for local languages comes at a time when generative AI is expanding the internet’s linguistic disparity and leaving behind thousands of languages.
Voice AI for 10 local languages in Africa and Asia
The lack of training data for low-resourced languages has long hindered multilingual customer experience automation. With large language models (LLMs) predominantly trained on common languages like English, Proto addressed this disparity by leveraging its ProtoAI natural language understanding engine for text-based communication. These solutions were optimised for mixed and local languages in Southeast Asia and Sub-Saharan Africa.
However, voice recognition posed a greater challenge than text, namely because of high-quality training data scarcity. Under the ScaleAI program, Proto’s consortium partners, including fellow Canadian company, Voices.com, are contributing 500 hours of spoken recordings for the prioritised languages. This data, combined with public-domain datasets and synthetic speech generation, underpin three key deliverables:
- Automated speech recognition: Accurate transcription of spoken audio in low-resourced languages.
- Text-to-speech capability: Synthesised responses that mimic natural human speech patterns.
- LLM training: Integration of historical chat records and contextual documents to enhance AI accuracy.
Immediately deployed for voice AI are Tagalog and Cebuano, the two dominant languages of the Philippines. In parallel to this local language development, Proto AI assistants can also use LLMs for voice responses in common languages like English and French, ensuring high-quality, natural interactions for customers that switch seamlessly between languages.
Proto CTO Weiying Kok said: “We’re grateful to our partners for supporting Proto’s expansion into voice AI for local languages. Working together, we’re overcoming the data availability challenge for these underserved languages and reaching even more customers.”
Scaling consumer protection for governments and enterprises
Proto’s voice AI launch is poised to make a significant impact on global consumer protection initiatives. In the Philippines, the technology will expand from serving five government agencies to over 30, supported by the country’s recent consumer protection laws such as the Anti Financial Account Scams Act, the Internet Transactions Act, and the No Wrong Door Policy enforced by the Office of the President. This initiative focuses on the five most spoken languages of the Philippines — Tagalog, Cebuano, Ilocano, Hiligaynon, and Bicolano — ensuring citizens can access services in their preferred language.
Gates Foundation Senior Program Officer Anna Wallace said: “The Foundation is investing in consumer protection AI tools with the goal of building trust in digital services for the lowest-income communities. Through these projects, we hope to demonstrate the value of an automated grievance redress system for citizens, regulators, and governments — and catalyse broader adoption across emerging markets”
Proto is also advancing consumer protection automation in sub-Saharan Africa, in collaboration with the Gates Foundation. The integration of voice AI technology for the current implementing countries of Liberia, Namibia and Rwanda will enhance the automation of complaint processing and other citizen support functions.
Giving voice to diverse customer experiences
Beyond consumer protection, Proto’s voice AI capabilities will serve a diverse range of use cases, from chat banking in Nigeria to remittance processing in Canada. The text-to-speech capacity will also support employee training and knowledge sharing, enabling organisations to engage more accessibly with diverse workforces.
