QContact is proud to announce the latest release of our market leading transcription engine, featuring improvements in both transcription and translation accuracy.
Since launching our own internal transcription engine earlier this year, we have transcribed hundreds of thousands of hours of real vernacular speech for our customers at no cost.
Unlike many of our competitors, we do not use your private recordings to train our transcription engine – instead we have built our own private dataset of thousands of hours of audio in each of the eleven languages of South Africa.
As we continue to grow this dataset, it has allowed us to improve our existing model and we’re proud to announce the latest release of our transcription engine featuring even better accuracy.
| Language | Old WER | WER | Change from January |
| Afrikaans | 11.0% | 6.1% | 45% |
| isiNdebele | 12.9% | 11.7% | 9% |
| Sepedi | 7.8% | 6.9% | 12% |
| Sesotho | 10.3% | 8.1% | 21% |
| siSwati | 13.2% | 13.2% | <1% |
| Setswana | 7.7% | 6.5% | 15% |
| Xitsonga | 11.0% | 8.5% | 23% |
| Tshivenda | 12.5% | 10.7% | 14% |
| isiXhosa | 19.1% | 13.3% | 30% |
| isiZulu | 20.6% | 15.7% | 24% |
We’re particularly proud of our improvements in Afrikaans, reducing transcription errors by over 40% and our improvements to isiXhosa and isiZulu where we have significantly reduced errors by over 20% on both languages.
Many transcription engines try to detect the language, then use a specialised model just for that language. That unfortunately doesn’t work in South Africa – most conversations contain code-switched language – a mix of languages, not just between sentences but the same sentence can contain multiple different languages or loan words.
QContact’s own proprietary model has been trained specifically to handle code switched languages, allowing customers and staff alike to switch between languages seamlessly without the transcription losing a beat.
What’s more, QContact continue to offer 100% call transcription free-of-charge to all our customers – saving you thousands every month compared to our competitors.
Read more in depth about our engine in our Call Transcription White Paper
