โ† All examples

ModelEthio-ASR Amharic (600M)8Ethio-ASR Multi (300M)14Ethio-ASR Multi (600M)9Ethio-ASR Multi (94M)9Meta MMS-1B9OmniASR-CTC (1B)9OmniASR-CTC (300M)9OmniASR-CTC (3B)7OmniASR-LLM (1B)7OmniASR-LLM (300M)8OmniASR-LLM (3B)7OpenAI Whisper-small16
Audio qualityClean Studio12Ambient Room Noise19Harsh Environment9Distant / Low-Volume Mic30Wideband Phone (VoLTE)9Randomized Phone Channel9Narrowband 2G + Packet Loss16

What did the model get wrong?

OpenAI Whisper-small Narrowband 2G + Packet Loss   Example 6 of 20  ยท  speaker: Male

๐ŸŽง Listen โ€” Narrowband 2G + Packet Loss
Heavy background noise plus phone-call compression โ€” worst-case stress test.
โœ… What was actually said
แˆ˜แ‰ธ แŠแ‹ แ‹จแˆšแˆตแ‰ฐแŠซแŠจแˆˆแ‹ แˆตแˆซ แˆตแˆ‹แˆˆแ‰ฅแŠ แ‰ถแˆŽ แ‰ขแˆ†แŠ•แˆแŠ แŒฅแˆฉ แŠแ‹
๐Ÿค– What this model heard
m akiyonan m is t arakalak t arakalak t alak t olok yon elu t olok
Correct
Wrong (different word)
Missed (skipped)
Extra (added)
0
Correct
9
Wrong
0
Missed
7
Extra
0%
Words right
Out of 9 spoken words, this model got 0 right (0%). Errors: 9 words wrong; 7 extra.

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