Technology
Fine-tuned automatic speech recognition for African languages—starting with Kiswahili customer experience (CX)—with measurable gains over out-of-the-box baselines.
How it works
Data collection
We capture real-world speech and metadata aligned with your domain—contact centres, BPO, and telephony—so models learn from audio that matches production noise, codecs, and accents.
Fine-tuning
Starting from strong open baselines, we adapt checkpoints to your languages and acoustic conditions. Evaluation uses normalized references and WER so improvements are measurable, not anecdotal.
Deployment
Tuned models integrate behind your existing APIs and workflows. You control rollout: compare baseline vs tuned in the same request when you need side-by-side assurance.
Continuous improvement
New clips, validator feedback, and production traffic feed the next training cycle—closing the loop from capture to better transcripts over time.
Accuracy
On our internal Kiswahili test set, a recent tuned checkpoint reduced word error rate from 50.70% (baseline) to 46.94% (tuned)—a relative improvement of about 7.4%. Figures depend on data and domain; we report them as internal benchmarks, not a universal guarantee for every deployment.
Supported languages
Kiswahili is the primary focus today. Our roadmap includes additional East African low-resource languages as we expand data partnerships and evaluation coverage.