Anufo Speech to Text: Preserving a Language Through AI

Anufo, also known as Chakosi, is a vibrant Kwa language spoken by the Anufo people across Ghana, Côte d'Ivoire, and Togo. With an estimated 200,000 speakers, it serves as a cornerstone of cultural identity, carrying centuries of oral history, proverbs, and song. However, like many minority languages, Anufo faces the risk of digital marginalisation. Without reliable speech-to-text tools, speakers cannot easily transcribe their spoken word into written form, limiting access to education, media, and online communication.

Why Accurate Speech-to-Text for Anufo Matters

Accurate transcription is essential for several reasons:

• Preservation of oral traditions: Elders' stories and songs can be recorded and transcribed for future generations.
• Education: Teachers can create reading materials in Anufo, helping children learn to read in their mother tongue.
• Media production: Podcasters, YouTubers, and community radio stations can subtitle their Anufo content, reaching wider audiences.
• Linguistic research: Linguists studying Anufo's grammar and phonology need high-quality transcriptions.
• Accessibility: Deaf or hard-of-hearing Anufo speakers can access video content through subtitles.

Challenges in Transcribing Anufo

Transcribing Anufo automatically is not trivial. The language uses three distinct tones (high, mid, low) that can change the meaning of a word. For example, “kɔ́” (high) means “to go”, while “kɔ̀” (low) means “to break”. Most generic ASR models ignore tone, leading to frequent errors. Additionally, there is very little publicly available training data for Anufo. Dialectal differences between the varieties spoken in Ghana, Côte d'Ivoire, and Togo also pose a challenge, as vocabulary and pronunciation can vary. Finally, the orthography is not fully standardised, with some writers using tone marks and others omitting them.

How Speechyou Overcomes These Challenges

Speechyou has developed a dedicated Anufo speech-to-text system that tackles these issues head-on. Our acoustic model is trained on a custom dataset of Anufo speech collected from native speakers across all three countries. We incorporate tonal analysis to accurately capture the three tones, using tone marks in the output by default. Separate dialect models ensure that the system recognises the unique features of Ghanaian, Ivorian, and Togolese Anufo. The output is normalised to a consistent orthography, but users can choose to customise it if needed.

Use Cases in Action

Anufo community organisations are already using Speechyou to preserve oral histories. For example, a group in the Volta Region recorded interviews with elders about traditional medicine and used Speechyou to transcribe them. The transcriptions were then published in a bilingual book (Anufo and English) for schools. In Côte d'Ivoire, a local filmmaker used Speechyou to generate Anufo subtitles for a documentary on farming techniques, making it accessible to the community. In Togo, a church transcribed weekly sermons into Anufo for distribution to members who prefer reading.

Getting Started with Anufo Transcription

Using Speechyou is straightforward. Simply upload your audio or video file, select “Anufo” as the language, and choose your dialect if known. The system will transcribe in real time or process the file quickly. You can then export the text as a plain transcript, SRT, or VTT subtitle file. The Unlimited plan is ideal for ongoing projects, allowing you to transcribe as much Anufo content as you need at no extra cost.

Conclusion

Anufo is a language rich in culture and history, and it deserves to be represented in the digital world. With Speechyou, speakers and researchers can now confidently transcribe Anufo speech, enabling preservation, education, and communication. Whether you are recording an oral history, creating subtitles, or developing language learning materials, Speechyou provides the accuracy and flexibility you need. Start your Anufo transcription project today and help keep the language alive for future generations.