Summary of Semantically Corrected Amharic Automatic Speech Recognition, by Samuael Adnew and Paul Pu Liang
Semantically Corrected Amharic Automatic Speech Recognition
by Samuael Adnew, Paul Pu Liang
First submitted to arxiv on: 20 Apr 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI); Machine Learning (cs.LG); Audio and Speech Processing (eess.AS)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary This paper focuses on enhancing Automatic Speech Recognition (ASR) for Amharic, a language spoken by over 50 million people primarily in eastern Africa. The authors build ASR tools for Amharic, which is written in the Ge’ez script with spacings that significantly impact sentence meaning. They find existing benchmarks do not account for these spacings and introduce a post-processing approach using a transformer encoder-decoder architecture to organize raw ASR outputs into grammatically complete sentences. The model enhances semantic correctness of Amharic speech recognition systems, achieving a Character Error Rate (CER) of 5.5% and a Word Error Rate (WER) of 23.3%. The authors release corrected transcriptions of existing Amharic ASR test datasets to enable accurate evaluation. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper helps make spoken languages more accessible worldwide by improving Automatic Speech Recognition (ASR) for Amharic, a language with over 50 million speakers in eastern Africa. Writing systems can be tricky! In this case, Amharic uses the Ge’ez script with special spaces between letters that matter when forming sentences. The authors noticed that current benchmarks don’t take these spaces into account, which makes it hard to measure how well ASR systems work in real-life situations. To fix this, they created a new approach using machine learning tools and tested it on existing datasets. This improved ASR system can understand Amharic speech better, making it more useful for people who rely on this language. |
Keywords
» Artificial intelligence » Cer » Encoder decoder » Machine learning » Transformer
