โ† All examples

ModelEthio-ASR Amharic (600M)2Ethio-ASR Multi (300M)6Ethio-ASR Multi (600M)6Ethio-ASR Multi (94M)7Meta MMS-1B10OmniASR-CTC (1B)4OmniASR-CTC (300M)2OmniASR-CTC (3B)2OmniASR-LLM (1B)2OmniASR-LLM (300M)6OmniASR-LLM (3B)2OpenAI Whisper-small15
Audio qualityClean Studio15Ambient Room Noise17Harsh Environment10Distant / Low-Volume Mic10Wideband Phone (VoLTE)10Randomized Phone Channel10Narrowband 2G + Packet Loss15

What did the model get wrong?

OpenAI Whisper-small Clean Studio   Example 12 of 20  ยท  speaker: Female

๐ŸŽง Listen โ€” Clean Studio
Clear, studio-quality recording. Best-case audio.
โœ… What was actually said
แŠ แˆแŠ• แ‹ซแˆˆแˆแ‰ต แŠ แ‹ณแˆ› แŠแ‹ แ‹ˆแ‹ฐ แŒ…แ‰กแ‰ฒ แˆˆแˆตแˆซ แˆแˆ„แ‹ต แˆตแˆˆแˆ†แŠ แŠแ‹
๐Ÿค– What this model heard
aounan lout had a dameano wadah dhada manu wadah djubutile sradhi tlwad jubutile sradhi tlounan
Correct
Wrong (different word)
Missed (skipped)
Extra (added)
0
Correct
10
Wrong
0
Missed
5
Extra
0%
Words right
Out of 10 spoken words, this model got 0 right (0%). Errors: 10 words wrong; 5 extra.

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