Amharic Speech Corpus: A 20-Hour Multi-Speaker Dataset for Automatic Speech Recognition

This dataset is a 20.03-hour Amharic speech corpus recorded from 100 native speakers and containing 10,850 carefully produced sentences with time-aligned transcriptions. Recordings are 16 kHz \/ 16-bit WAV files, manually segmented and verified for transcription consistency. The prompts were chosen to give broad phonetic and lexical coverage, making the corpus useful for building and evaluating Amharic ASR systems, acoustic and language models, and linguistic studies of the language.