Artificial intelligence (AI) is undoubtedly advancing quickly; from ChatGPT-3.5, which astounded us in 2022, to Sora and GPT-4o, which have left people perplexed, there’s no denying it.
While AI continues to advance, it’s safe to conclude that Africa has yet to advance. Even though several nations, like Nigeria, have developed programmes and regulations to further AI development, the outcomes could have been more impressive.
“The first Igbo voice-to-text AI model that resonates with Africans is called IgboSpeech, created by Ijemma Onwuzulike, born in the United States.”
Motivation led Ijemma to learn and research languages, discover patterns and processes in language learning, and understand the process of compiling data on platforms like Duolingo.
Read also: OpenAI unveils GPT-4o Mini: A new era in cost-efficient AI
Innovations and Challenges in the Making of the Igbo Digital Dictionary
Duolingo is just a language learning programme; much data entry, research, and linguistic labour goes into creating that user-friendly interface behind the scenes.
Onwuzulike decided to develop an Igbo API in 2020 to make it simpler for a website like Duolingo to include Igbo as one of its languages.
The API provides a digital dictionary for Igbo language developers, enabling easy retrieval of word-related data like definitions, synonyms, antonyms, and usage examples.
Onwuzulike and her team faced challenges building a word API due to inconsistencies in existing Igbo dictionaries and ensuring a quick and efficient response to user requests.
The Igbo API, which includes 5,119 words, 29,922 sentences, audio recordings, 571 definitions, and 464 proverbs, is being developed with a robust backend, caching, and growth capabilities.
She pursued Computer Science at Dartmouth College, studying Japanese and literature, and later worked in machine learning and artificial intelligence. They interned twice at eBay and later worked for Squarespace.
Read also: Anambra University partners tech firm to empower students with AI
IgboSpeech: Unlocking Voice-to-Text Translation for Igbo Language
Launched on July 1, 2024, is the demo website for IgboSpeech. Igbo speech and text are catered for in this automated speech recognition (ASR) model.
In addition, it uses the Igbo API in part. That does, however, have a unique dataset. According to the website, it records Igbo speeches and converts them into text.
Onwuzulike plans to use IgboSpeech to automatically generate subtitles for Igbo movies, YouTube videos, and note-taking apps, focusing on voice-to-text translation instead of text-to-voice.
Onwuzulike and her team aim to improve IgboSpeech’s features through grants, as the project is non-profit and currently offers the best value for money for translators and text-to-voice translations. They are examining the Lacuna Fund, which provides funding for developing natural language processing (NLP) and antimicrobial resistance (AMR).