Fair Forward - Open Data & Use Cases

This dataset and the models developed from it aim to predict biomass levels in shaded regions of Côte d'Ivoire using a combination of GEDI, Sentinel-2 satellite imagery, and ground truth biomass measurements. The resulting models enable efficient and cost-effective remote monitoring of biomass changes, crucial for assessing reforestation success and detecting forest degradation due to cocoa farming, reducing the need for extensive on-the-ground surveys.

This speech dataset for the Ghanian languages Akan (Akuapem Twi, Asante Twi, Fante) and Ga includes 104,000 utterances (speech) across the four dialects/languages with approximately 200 speakers per dialect/language. This amounts to about 148 hours of speech in total. The dataset was developed to support the development of financial applications in native Ghanaian languages to allow illiterate and semi-literate people to fully benefit from digital financial services. Secondly, it aims to answer research questions related to domain-specific vs. general-purpose dataset development, dialects, as well as NLP system development in low resource settings.

This dataset supports yield estimation, crop type detection and classification, fruit detection and counting, and fruit maturity stage detection (unripe, ripe, and spoiled) for three products that are important sources of livelihood for millions of households in Sub-Saharan Africa. It contains 14,870 drone images with bounding box annotations of cashew, cocoa, and coffee trees collected across multiple farms in Ghana and Uganda. Conventional methods of yield estimation are expensive, require a lot of labor and time, and are prone to error due to incomplete ground observations. This results in poor crop yield estimations and hinders farmers’ ability to appropriately plan and manage their fields and production pipelines. This dataset will help transform African agriculture into agribusiness by allowing for the development of yield estimation solutions that enable farmers to make good business decisions. Having key details about agricultural production readily accessible enables a timely harvest, helping farmers ensure healthy, fresh produce and, in addition, better sales.

The Accent Classification Dataset (Ghana) is a collection of audio recordings from native and non-native English speakers across Ghana's diverse regions. Participants read the same three scripts, capturing distinct regional accents and speech patterns. This consistent dataset is valuable for linguistic analysis, accent classification, and speech recognition research focused on Ghanaian English.

The Agrivoltaic system offers a transformative solution for farming communities by providing a means to generate electricity without sacrificing agricultural productivity. The dataset showcases the effectivenesss of Solar PVs and also crop yield under solar PVs and in the open-sun on the same farm.

This dataset is part of the initiative SYSPIN (SYnthesizing SPeech in INdian languages), that develops large open-source Text-to-Speech (TTS) corpora, i.e., speech training datasets and AI models in 9 Indian languages, namely – Bengali, Bhojpuri, Chhattisgarhi, Hindi, Kannada, Magahi, Maithili, Marathi, and Telugu. This data is useful for those wishing to build voice-enabled, user-friendly applications in Indian languages. Through voice-based solutions, these AI models can be used to empower communities that do not have access to digital services in their own native language. They can be used by private companies, researchers, NGOs or government departments to develop digital applications. The datasets and TTS models are available to the AI community and anyone wishing to build voice-enabled applications in Indian languages.

100 hours of Text to Speech Dataset for Mundari Language

This project focuses on creating validated, open-source High Carbon Stock (HCS) forest maps for Indonesia using an AI-driven approach. It adopts the HCS classification system to identify forests suitable for conservation versus sustainable development: • High, Medium, and Low-Density Forests (Conservation Priority): Forests with closed or patchy canopies, rich biodiversity, and carbon storage. • Young Regenerating Forests: Regrowing forests that are transitioning into higher carbon stock potential. • Scrub and Open Land: Areas with minimal carbon stock suitable for sustainable development. The maps integrate field data, satellite imagery, and AI to assist policymakers, NGOs, and businesses in making data-driven decisions for sustainable land-use planning and conservation. Stakeholders can use the outputs for spatial planning, reducing greenhouse gas emissions, and implementing "no deforestation" commitments. The High Carbon Stock Approach (HCSA) project was nominated for the prestigious UN SDG Game Changer award – Category 2: Planet – Innovation for our climate and environment “Artificial Intelligence Meets Jungle/ Forest Forward.

By collecting more than 1000 hours of AI voice data (by 04/2025), this effort created the largest open-source voice dataset of diverse Swahili speakers for speech recognition (speech-to-text). You can use this resource to build AI systems understanding spoken Swahili e.g. in cellphones . This facilitates access to information in mother tongue, including by illiterate persons.

Developing products that can be replicated through four initiatives: 1. DynAG: Focused on rice, wheat, and maize, DynAG's platform uses a RAG pipeline powered by LLMs like the GPT suite to provide farmers with advisory services. Accessible via chatbots, Android and Jio-based mobile apps, IVR, and SMS, it ensures wide reach. With machine translation and speech recognition, the platform supports communication in English and Hindi while utilizing CGIAR's data backbone for accurate, data-driven insights. 2. Digital Green: Digital Green's platform supports agricultural value chains like dairy, coffee, and wheat in Bihar and Kenya. It integrates a RAG pipeline, using LLMs under evaluation, and offers services through an app, Telegram, and IVR. Features like machine translation, speech technology, image recognition, and weather data enable localized insights. The platform supports multiple languages, including Hindi, Swahili, Gikuyu, and Marathi, ensuring flexibility across regions. 3. Viamo and Partners: This platform targets farmers in Kenya and Bihar, focusing on beans and wheat, respectively. Using IVR, SMS, and WhatsApp, the platform is built on a RAG pipeline powered by LLMs, with potential for intent and entity recognition. Machine translation and speech-enabled IVR support English, Hindi, and Swahili, making it accessible across regions. The initiative is part of a broader plan to enhance voice technology for agricultural advisory. 4. Tech for Her: Tech for Her empowers women farmers in Bihar and Kenya, focusing on tomatoes and cattle. Accessible via IVR, WhatsApp, and a mobile app, the platform uses LLM-powered intent and entity recognition for personalized chatbot interactions and speech-enabled IVR for voice support. AI-based video generation further enhances engagement. Available in English, Hindi, and Swahili, it caters to the specific needs of women in agriculture.

This team collected parallel text corpora for three Kenyan indigenous languages, Kidaw’ida, Kalenjin, and Dholuo, alongside Kiswahili, resulting in approximately 90,000 sentence pairs in total. After collection, the team separated out the Kidaw’ida, Kalenjin, and Dholuo sentences and used them as monolingual datasets for crowd-sourcing speech data, facilitated by uploading the sentences to Mozilla Common Voice. A total of 109 members of the three language communities were recruited to read and record sentences from their respective native languages. Emphasizing gender balance and including different ages and regional variants helped to make the datasets more representative. The voice datasets offer a substantial amount of speech data, comprising 56 hours of Kidaw’ida, 92 hours of Kalenjin, and 120 hours of Dholuo, for a total of 268 hours. Use cases for these parallel corpora include training models to translate text between Kiswahili and Kidaw’ida, Kalenjin, and Dholuo. The speech data on Mozilla Common Voice, along with its associated text data, is intended to be used for the development of speech recognition applications. The languages that comprise this dataset are low-resource, especially Kidaw’ida, which has only around 400,000 speakers and faces a more immediate risk of loss. By collecting the text and speech data, this team contributed to the preservation of these languages. They hope that once enough data has been collected to train accurate models and create NLP applications for these three languages, they will become more relevant in the modern digital age, thus mitigating the risk of loss.

This team annotated 2,125 Google Earth satellite images and 9,202 drone images, forming a combination of low and high-definition solar panel views in Madagascar. The Madagascar Initiatives for Digital Innovation (MAIDI) team performed field checks for up to 25% of satellite images and, in total, annotated 22,488 polygons. This dataset will help data scientists and users develop a solar panel detection algorithm to measure solar energy adoption across Madagascar. Notably, this project represented all regions of the country; instead of focusing only on big cities, it also covered average and small villages as well as coasts and mountains.

This dataset is the first large-scale human-annotated Twitter sentiment dataset for Hausa, Igbo, Nigerian-Pidgin, and Yorùbá, the four most widely spoken languages in Nigeria.

This effort created the largest open-source voice dataset of diverse Kinyarwanda speakers for speech recognition (speech-to-text). It collected more than 2300 hours of AI voice data in Kinyarwanda (by 04/2025). You can use this resource to build AI systems understanding spoken Kinyarwanda e.g. in cellphones . This facilitates access to information in mother tongue, including by illiterate persons.

Field data collected on agricultural farm land (crops) with raw datasets + applied use cases of classification of land use cover using ML (based on individual projects presented)

Univ of Stellenbosch Grant (Febe de Wet) - Voices of Mzansi project / localise SA local languages on Mozilla CV platform

This dataset was produced as part of the Radiant Earth Spot the Crop Challenge (https://zindi.africa/hackathons/radiant-earth-spot-the-crop-hackathon). The objective of the competition was to create a machine learning model to classify fields by crop type from images collected during the growing season by the Sentinel-2 and Sentinel-1 satellites.

This dataset contains annotated sentences with personally identifiable information (PII) in Luganda, Lumasaba, Hausa, and Kanuri. These four languages span Central and Eastern Uganda, Nigeria, Ghana, and Northern Cameroon. The team collected 3,000 sentences for both Kanuri and Hausa, 5,000 for Lumasaba, and 4,000 for Luganda. Potential use cases for these datasets include named entity recognition (NER), text classification, privacy-preserving data analysis and research, language modeling, machine translation, and linguistic research. The team aimed to curate a dataset that is gender inclusive, and their work highlighted the need for standardized guidelines for annotating low-resourced languages. Having these guidelines would help to avoid common pitfalls and errors when labeling text data in these low-resource languages.