โ† All examples

ModelEthio-ASR Amharic (600M)13Ethio-ASR Multi (300M)20Ethio-ASR Multi (600M)12Ethio-ASR Multi (94M)16Meta MMS-1B22OmniASR-CTC (1B)5OmniASR-CTC (300M)13OmniASR-CTC (3B)10OmniASR-LLM (1B)8OmniASR-LLM (300M)6OmniASR-LLM (3B)9OpenAI Whisper-small22
Audio qualityClean Studio14Ambient Room Noise5Harsh Environment7Distant / Low-Volume Mic13Wideband Phone (VoLTE)5Randomized Phone Channel5Narrowband 2G + Packet Loss13

What did the model get wrong?

OmniASR-CTC (300M) Narrowband 2G + Packet Loss   Example 2 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
แ‰ฃแˆ…แˆญ แ‹ณแˆญ แАแ‹ แ‹ซแˆˆแˆแ‰ต แŠขแŠ•แ‰ฐแˆญแŠ”แ‰ต แ‰ฐแ‰‹แˆญแŒฆแ‰ฅแŠ›แˆ แ‰ฅแˆแˆฝแ‰ต แˆ›แˆตแˆ˜แ‹แŒˆแ‰ฅ แˆแˆแŒŒ แАแ‰ แˆญ แˆตแˆแŠฌ แ‹œแˆฎ แ‹˜แŒ แŠ แŠ แŠ•แ‹ต แŠ แŠ•แ‹ต แˆตแ‹ตแˆตแ‰ต แˆฐแ‰ฃแ‰ต แ‹œแˆฎ แˆแˆˆแ‰ต แ‹˜แŒ แŠ แŠ แˆแˆตแ‰ต แАแ‹
๐Ÿค– What this model heard
แ‰  แŠ แ‹ซแˆญแА แ‹ซแˆˆแ‹แ‰ต แ‰ดแˆŒแ‰จแˆญ แ‰ฐแ‰‹แˆญแŒฆแ‰ฅแŠ›แˆ แ•แˆแˆฝแ‰ฅ แˆ›แˆตแАแ‰ต แˆˆแˆแˆแŒ แ‹จแАแ‰ แˆญ แˆตแˆแ‰„ แ‹œแˆฎ แ‹˜แŒ แŠ แŠ แŠ•แ‹ต แŠ แŠ•แ‹ต แˆตแŒแˆตแ‰ต แˆฐแ‰ฃแ‰ต แ‹œแˆฎ แˆแˆˆแ‰ต แ‹ˆแŒ แŠ แŠ แˆแ‰ต แАแ‹
Correct
Wrong (different word)
Missed (skipped)
Extra (added)
9
Correct
12
Wrong
1
Missed
0
Extra
41%
Words right
Out of 22 spoken words, this model got 9 right (41%). Errors: 12 words wrong; 1 missed.

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