DONDO: Open w2v-BERT Speech-Recognition Base Models for African Languages

2026-07-23 · Source: Takara TLDR - Daily AI Papers · Field: Technology & Digital — Artificial Intelligence & Machine Learning · Depth: Expert, quick

Summary

DONDO is a new family of open, permissively licensed automatic speech recognition (ASR) base models specifically designed for African languages. Built upon the w2v-BERT 2.0 self-supervised speech encoder, DONDO includes twenty-one monolingual and five multilingual models, covering twenty-seven language varieties across Ghana, Sierra Leone, Nigeria, Senegal, Kenya, and Zimbabwe. These models are fine-tuned using a two-step (or three-step for one family) learning-rate-annealed procedure, primarily on read speech from religious texts to ensure broad, license-clear, and orthographically consistent coverage. A lightweight language-conditioning mechanism allows multilingual checkpoints to be steered to a target language at inference by injecting a one-hot language identity. The annealed multilingual models achieve average word error rates (WER) of 10-13%, significantly closing the gap to monolingual baselines. All models are released under the Apache-2.0 license on Hugging Face KhayaAI, supporting free commercial use for languages spoken by an estimated one hundred million first-language speakers.

Key takeaway

For Machine Learning Engineers developing ASR solutions for African languages, DONDO provides a robust, open-source foundation. You can utilize these w2v-BERT 2.0 models, available under Apache-2.0, to accelerate development and deployment. The fine-tuning procedure and language-conditioning mechanism offer practical strategies for adapting multilingual models. Consider integrating these models to serve the estimated one hundred million first-language speakers, reducing reliance on proprietary or less performant alternatives.

Key insights

Open w2v-BERT 2.0 ASR models for African languages achieve strong performance with a novel fine-tuning and language-conditioning approach.

Principles

• Religious texts offer license-clear data for low-resource languages.
• Annealed fine-tuning can recover monolingual baseline performance.
• Language conditioning enables multilingual model steering.

Method

A two-step (or three-step) learning-rate-annealed fine-tuning procedure adapts a shared multilingual model, followed by annealing. Language conditioning injects a one-hot language identity as prefix frames.

In practice

• Fine-tune DONDO models for specific African language ASR tasks.
• Utilize the language-conditioning mechanism for multilingual inference.
• Explore religious texts as a data source for other low-resource languages.

Topics

• Automatic Speech Recognition
• African Languages
• w2v-BERT 2.0
• Self-supervised Learning
• Multilingual Models
• Low-resource Languages
• Hugging Face KhayaAI

Best for: AI Engineer, Research Scientist, AI Scientist, Machine Learning Engineer, NLP Engineer

Related on AIssential

• DONDO: Open w2v-BERT Speech-Recognition Base Models for African Languages — Open w2v-BERT 2.0 ASR base models for 27 African languages achieve competitive WERs via annealed fine-tuning and prefix-frame conditioning.
• DONDO: Open w2v-BERT Speech-Recognition Base Models for African Languages — Open w2v-BERT ASR models for African languages achieve strong performance via annealed fine-tuning and language conditioning.
• Voices from the Margins: Modeling Linguistic Diversity in Spontaneous Speech for Low-Resource Languages — ASR for low-resource languages benefits from multilingual models and n-gram decoding, aiding revitalization and accessibility.
• Expanding the Lexicon of Ge'ez Based African Languages: A Comparative Study of Amharic and Tigrinya — Vocabulary extension and two-stage training significantly boost multilingual PLM performance on low-resource, non-Latin-script languages.
• Optimization of Voice Translation Systems for Indigenous Languages: Retraining the NLLB-200 Model for the Quechua–Spanish Pair — Fine-tuning NLLB-200 with curated data and optimized training improves low-resource indigenous language translation on consumer hardware.
• From Sinhala to Dhivehi: Cross-Lingual Transfer Learning for Low-Resource Speech Recognition — Cross-lingual transfer for low-resource ASR benefits from linguistic relatedness and specific adaptation strategies, with external language models being critical.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Takara TLDR - Daily AI Papers.