We benchmarked Addis Voice 2 against six other text-to-speech systems on Amharic.
100 prompts from FLEURS, Waxal and Horn-ASR. Seven systems. Same text across all of them.
We measured it two ways: how it sounds to native Amharic listeners, and how accurately a third-party speech recogniser transcribes it.
Results: https://addisassistant.com/benchmarks/addis-voice-2-amharic
Dataset: huggingface.co/datasets/addisai/amharic-tts-benchmark
@addisassistantai
