Amharic speech recognition struggles with phone call noise and gets better with training
TeleASR-Amharic: A Benchmark for Robust Amharic ASR in Realistic Telecommunication Conditions
Computation and LanguageSound
Summary
Speech recognition works poorly for Amharic when used on real phone calls because calls add lots of noise and distortions. The authors studied these problems and created a set of tests using real phone call sounds to see how well Amharic speech models perform. They found that even top models make many more mistakes on noisy calls but can improve if trained on data that mimics these conditions. They also released this phone-call-style data so others can test and improve Amharic speech recognition systems.
What this means in practice
- •For telecom software developers: Improve Amharic voice transcription accuracy by using telephony noise-aware training data included in this benchmark.
- •For call center technology teams: Use the benchmark data to evaluate and enhance Amharic speech recognition systems dealing with noisy customer calls.
- •For multilingual voice assistant makers: Develop more reliable Amharic speech input for assistants by training models on varied telecom noise conditions described here.$Commercial implications: Enables selling more accurate Amharic speech recognition features for voice assistants used in noisy phone environments.
Authors
Dagnachew Mekonnen Marilign, Enrique Molero, Abrham Haile, Louis Powell, Israel Abebe Azime
Abstract
The real-world usability of ASR models remains challenging, particularly for low-resource languages, where existing speech resources are often collected in clean studio environments. In this work, we introduce TeleASR-Amharic, a benchmark for evaluating Amharic ASR under various acoustic and telecommunication conditions. We analyze the major degradations observed in real telecommunications recordings and develop a benchmark that covers multiple channel conditions, environmental disturbances, and telecom-domain variations. Our results show that telecommunication degradation can substantially reduce recognition performance, making even high-performing ASR models impractical under severe conditions. For example, the WER of the best-performing model increases from 0.23 to 0.79 under severe telecommunication degradation, corresponding to a relative increase of 243% in recognition errors. We further investigate the effects of different noise and channel distortions and show that scaling training data and applying carefully designed telecommunication-aware augmentation substantially improve robustness. In particular, our best training configuration achieves a 23.6% relative WER reduction under extreme noise conditions. Together with the release of our telecommunications-domain data, these findings provide a reproducible benchmark for measuring progress in Amharic ASR while highlighting open challenges in developing robust speech technologies for real-world deployment.