โ† All examples

ModelEthio-ASR Amharic (600M)5Ethio-ASR Multi (300M)16Ethio-ASR Multi (600M)13Ethio-ASR Multi (94M)12Meta MMS-1B13OmniASR-CTC (1B)13OmniASR-CTC (300M)13OmniASR-CTC (3B)13OmniASR-LLM (1B)13OmniASR-LLM (300M)12OmniASR-LLM (3B)12OpenAI Whisper-small14
Audio qualityClean Studio5Ambient Room Noise5Harsh Environment6Distant / Low-Volume Mic5Wideband Phone (VoLTE)6Randomized Phone Channel7Narrowband 2G + Packet Loss12

What did the model get wrong?

Ethio-ASR Multi (94M) Narrowband 2G + Packet Loss   Example 13 of 20  ยท  speaker: Female

๐ŸŽง Listen โ€” Narrowband 2G + Packet Loss
Heavy background noise plus phone-call compression โ€” worst-case stress test.
โœ… What was actually said
แ‹ซแŠ›แ‹ แˆตแˆแŠญ แˆŒแˆ‹ แˆ˜แˆณแˆชแ‹ซ แˆ‹แ‹ญ แŠแ‹ แ‹ซแˆˆแ‹ แŠจแ‹šแˆ…แŠ›แ‹ แ‰แŒฅแˆญ แˆ‹แ‹ญ แˆ†แŠ˜ แˆ˜แŒจแˆจแˆต แŠ แˆแ‰ฝแˆแˆ
๐Ÿค– What this model heard
tir แŠ“ แŠญแˆ‹ แˆ˜แˆณแˆชแ‹ซ แŠ“แ‹ญแŠ“ แˆ˜
Correct
Wrong (different word)
Missed (skipped)
Extra (added)
1
Correct
5
Wrong
7
Missed
0
Extra
8%
Words right
Out of 13 spoken words, this model got 1 right (8%). Errors: 5 words wrong; 7 missed.

Try switching Audio quality โ€” see the same model degrade as the audio gets noisier.