โ† All examples

ModelEthio-ASR Amharic (600M)2Ethio-ASR Multi (300M)7Ethio-ASR Multi (600M)4Ethio-ASR Multi (94M)8Meta MMS-1B16OmniASR-CTC (1B)6OmniASR-CTC (300M)9OmniASR-CTC (3B)8OmniASR-LLM (1B)10OmniASR-LLM (300M)12OmniASR-LLM (3B)10OpenAI Whisper-small16
Audio qualityClean Studio2Ambient Room Noise9Harsh Environment9Distant / Low-Volume Mic2Wideband Phone (VoLTE)10Randomized Phone Channel10Narrowband 2G + Packet Loss13

What did the model get wrong?

OmniASR-LLM (3B) Randomized Phone Channel   Example 17 of 20  ยท  speaker: Male

๐ŸŽง Listen โ€” Randomized Phone Channel
Phone-call audio with random extra distortions layered on top.
โœ… What was actually said
แ‰ แˆ˜แ‰ฐแŒแ‰ แˆชแ‹ซแ‹ แ‹แˆตแŒฅ แˆฎแˆšแŠ•แŒ แˆฐแˆญแ‰ชแˆต แ‹จแˆšแˆ แˆšแŠ’ แˆ˜แ‰ฐแŒแ‰ แˆชแ‹ซ แ‹ซแŒˆแŠ›แˆ‰ แŠฅแ‹› แ‹แˆตแŒฅ แŒˆแ‰ฅแ‰ฐแ‹ แŠ แŒˆแˆแŒแˆŽแ‰ฑแŠ• แˆ›แˆตแŒ€แˆ˜แˆญแˆ แˆ†แА แˆ›แ‰‹แˆจแŒฅ แ‹ญแ‰ฝแˆ‹แˆ‰
๐Ÿค– What this model heard
แ‰  แˆ˜แ‰ฐแŒแ‰ แˆชแ‹ซ แ‹แˆตแŒฅ แˆฎแˆšแŠ•แŒ แˆฐแˆญแ‰ชแˆต แ‹จแˆš แˆ แˆšแŠ’ แˆ˜แ‰ฐแŒแ‰ แˆชแ‹ซ แ‹ซแŒˆแŠ› แˆ‰ แŠฅแ‹šแ‹ซ แ‹แˆตแŒฅ แŒˆแ‰ฅแ‰ฐแ‹ แŠ แŒˆแˆแŒแˆŽแ‰ฑ แŠ• แˆ›แˆตแŒ€แˆ˜แˆญ แˆ†แА แˆ›แ‰‹แˆจแŒฅ แ‹ญแ‰ฝแˆ‹แˆ‰
Correct
Wrong (different word)
Missed (skipped)
Extra (added)
10
Correct
6
Wrong
0
Missed
4
Extra
62%
Words right
Out of 16 spoken words, this model got 10 right (62%). Errors: 6 words wrong; 4 extra.

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