ScaleAI, an innovation cluster dedicated to accelerating the adoption of Canadian artificial intelligence, has announced a grant of approximately US$500,000 into a Proto-led consortium to advance generative AI customer experiences in underserved languages.
Through the partnership with ScaleAI, Proto will upgrade the latest version of its AI Customer Experience (AICX) Platform with automated speech recognition capability for underserved languages such as Tagalog and Kinyarwanda. Such languages, referred to as low-resourced due to a lack of training data, have hitherto been unsupported in the rapidly growing generative AI customer experience industry.
These enhancements will localise and improve the performance of Proto’s AICX solutions for the government, health, transport, and finance sectors, in partnership with consortium contributors including Canadian firms such as Voices.com and SQL Power.
This voice AI capability will be deployed for enterprises and governments serving multilingual customers for functions such as complaint processing and remittance payment troubleshooting.
Voice AI for 10 local languages in Africa and Asia
The absence of training data for low-resourced languages is a major barrier to multilingual CX automation. This inequality is widened by large language models (LLMs), which are 93% trained on common languages like English.
To fill this gap, Proto augments LLM-powered chatbots with its ProtoAI™ natural language understanding engine, specialised in local and mixed languages in Southeast Asia and Sub-Saharan Africa. However, meaningful improvements in text accuracy and speech word error rates (WERs) require a massive volume of training data. Such data is difficult to generate due to poor translation quality, data privacy limitations, and noise interference.
To solve this, consortium partner, Voices.com, will deliver 500 hours of spoken recordings for prioritised languages, including rarer options like Hiligaynon and Nyanja. This contribution is key to the project’s three deliverables:
- Automated speech recognition with accurate transcription of spoken audio output for low-resourced languages from three types of data: public-domain voice datasets, 500 hours of high-quality audio recordings from Voices.com, and 500 hours of synthetic speech generation.
- Text to speech capability with synthesised responses generated from chat transcripts and the recordings from Voices.com. Proto will also leverage a public-domain speech dataset to ensure the output is natural-sounding, mimicking human speech patterns.
- LLM training with contextual data, including historical chat records and documents, in low-resourced languages.
Scaling consumer protection for government agencies
The highest-impact application of the advancements from this ScaleAI project is Proto's parallel scaleup initiative for its consumer protection usecase, starting in the Philippines with an expansion from 5 to 50 government agencies. This initiative, supported by the country’s new Internet Transactions Act, integrates government agency chatbots and complaints management systems to provide a unified, whole-of-government approach to consumer protection.
The ScaleAI project’s focus on the five most spoken languages of the Philippines — Tagalog, Cebuano, Ilocano, Hiligaynon, and Bicolano — is particularly significant for including the majority of citizens in their preferred spoken or written language.
This initiative builds on Proto’s current text-based deployments for government agencies such as the Central Bank of the Philippines, where 76% of financial sector complaints processing was automated. Following this launch, Proto and its implementation and fundings partner, including the University of Cambridge SupTech Lab, will expand this voice AI application into multiple sub-Saharan Africa countries.
Beyond this scaleup initiative for consumer protection in low-to-middle income countries, Proto will be working with consortium partner SQL Power to apply the AI-enabled consumer protection capability into developed markets.
Giving voice to customer and employee experiences
Proto chatbots also serve a diverse range of customer and employee experience usecases, from chat banking in Nigeria to contact centre agent assistance in China. Often, these client deployments use the same chatbots or data sources to serve both customers and internal employees in need of training and ongoing support. The development effort to upgrade these chatbots with voice AI will be spearheaded by Proto’s Head of Engineering, Weiying Kok.
Proto Head of Engineering, Weiying Kok, said, “We’re grateful to ScaleAI for supporting Proto and its consortium. This project overcomes the data availability challenge for underserved languages, and provides the opportunity to achieve exceptional AI accuracy rates, with exciting applications across enterprise and government.”
Within the Canadian consortium, the remittances platform, Manila Express, and indoor wayfinding vendor, Mappedin, are partnered with Proto to rollout the voice-enhanced chatbot capability to respective customer bases. The speech recognition advancement from the ScaleAI project will also be applied to Proto’s clients on a per-language basis once the targeted word-error rates are achieved.
About ScaleAI
ScaleAI is a Canadian innovation cluster dedicated to accelerating the adoption of artificial intelligence across various industries. As an independent, not-for-profit organisation, ScaleAI unites private enterprises, research institutions, and startups to build intelligent supply chains and enhance industry performance through AI-powered solutions. The organisation has facilitated the support of over 120 AI projects, amounting to investments of approximately $600 million, focusing on sectors such as healthcare, manufacturing, transportation, and information technology.