← All examples

ModelEthio-ASR Amharic (600M)5Ethio-ASR Multi (300M)1Ethio-ASR Multi (600M)3Ethio-ASR Multi (94M)1Meta MMS-1B11OmniASR-CTC (1B)0OmniASR-CTC (300M)6OmniASR-CTC (3B)0OmniASR-LLM (1B)0OmniASR-LLM (300M)0OmniASR-LLM (3B)0OpenAI Whisper-small11
Audio qualityClean Studio6Ambient Room Noise0Harsh Environment0Distant / Low-Volume Mic5Wideband Phone (VoLTE)1Randomized Phone Channel0Narrowband 2G + Packet Loss11

What did the model get wrong?

OmniASR-CTC (300M) Clean Studio   Example 16 of 20  ·  speaker: Male

🎧 Listen — Clean Studio
Clear, studio-quality recording. Best-case audio.
✅ What was actually said
ቁጥሩ ስድስት ሰባት ዘጠኝ ሶስት ሁለት ስድስት ሰባት ስድስት አራት ነው
🤖 What this model heard
ቁጥሩ ስስ ሰባት ዘጠ ሶስትስት ሁለት ስስ ሰት ስስ አራት ነው
Correct
Wrong (different word)
Missed (skipped)
Extra (added)
5
Correct
6
Wrong
0
Missed
0
Extra
45%
Words right
Out of 11 spoken words, this model got 5 right (45%). Errors: 6 words wrong.

Try switching Audio quality — see the same model degrade as the audio gets noisier.