Transcription in 98 languages
EKHOS AI runs on top of OpenAI Whisper, one of the most advanced transcription AI models available today. The Whisper model supports 98 languages. The table below lists its accuracy based on the large v3 model, which is used by Expert mode in EKHOS AI.
Note: accuracy ≈ 100 − WER/CER (%). CER (character error rate) values were treated the same for conversion.
The accuracy table was reference form Open AI Whisper
Common Voice 15 — large-v3 (Expert Accuracy) (%)
Language
Accuracy (%)
Dutch 95.7 Spanish 95.3 Korean 94.8 Italian 94.5 German 94.3 Thai 94.2 Russian 94.2 Portuguese 94.1 Polish 94.0 Indonesian 92.8 Mandarin (TW) 91.8 Swedish 91.7 Czech 91.0 English 90.7 Japanese 89.7 French 89.2 Romanian 89.2 Cantonese (CN) 89.1 Turkish 88.0 Mandarin (CN) 87.2 Catalan 86.7 Hungarian 86.6 Ukrainian 86.4 Greek 86.3 Bulgarian 85.7 Arabic 84.9 Serbian 84.3 Macedonian 84.2 Cantonese (HK) 84.1 Latvian 83.3 Slovenian 83.2 Hindi 83.1 Galician 82.0 Danish 81.9 Urdu 79.6 Slovak 76.6 Hebrew 76.5 Finnish 75.4 Azerbaijani 75.2 Lithuanian 72.4 Estonian 70.1 Nynorsk 69.3 Welsh 68.0 Punjabi 65.3 Afrikaans 64.0 Persian 62.6 Basque 61.1 Vietnamese 60.1 Bengali 59.7 Nepali 59.6 Marathi 58.8 Belarusian 56.3 Kazakh 52.4 Armenian 51.9 Swahili 48.8 Tamil 48.5 Albanian 44.3
FLEURS — large-v3 (Expert Accuracy) (%)
Language
Accuracy (%)
Spanish 97.2 Italian 97.0 Korean 96.9 Portuguese 95.9 English 95.9 Polish 95.4 Catalan 95.2 Japanese 95.1 German 95.1 Russian 95.0 Dutch 94.8 French 94.7 Indonesian 93.9 Ukrainian 93.6 Turkish 93.3 Malay 92.7 Swedish 92.4 Mandarin 92.3 Finnish 92.3 Norwegian 92.2 Romanian 91.8 Thai 91.6 Vietnamese 91.4 Slovak 90.8 Arabic 90.4 Czech 89.9 Croatian 89.2 Greek 89.1 Serbian 88.4 Danish 88.0 Bulgarian 87.5 Hungarian 87.1 Filipino 87.0 Bosnian 87.0 Galician 86.9 Macedonian 85.3 Hindi 83.0 Estonian 81.9 Slovenian 81.7 Tamil 81.7 Latvian 80.6 Azerbaijani 80.3 Urdu 79.4 Lithuanian 76.3 Hebrew 73.9 Welsh 71.4 Persian 70.6 Icelandic 69.6 Kazakh 67.6 Afrikaans 67.6 Kannada 67.0 Marathi 65.9 Swahili 65.9 Telugu 60.7 Maori 60.2 Nepali 59.8 Armenian 57.8 Belarusian 57.5 Gujarati 52.6 Punjabi 51.5 Bengali 50.0
Common Voice 15 — large-v3 (Expert Accuracy) (%)
| Language | Accuracy (%) |
|---|---|
| Dutch | 95.7 |
| Spanish | 95.3 |
| Korean | 94.8 |
| Italian | 94.5 |
| German | 94.3 |
| Thai | 94.2 |
| Russian | 94.2 |
| Portuguese | 94.1 |
| Polish | 94.0 |
| Indonesian | 92.8 |
| Mandarin (TW) | 91.8 |
| Swedish | 91.7 |
| Czech | 91.0 |
| English | 90.7 |
| Japanese | 89.7 |
| French | 89.2 |
| Romanian | 89.2 |
| Cantonese (CN) | 89.1 |
| Turkish | 88.0 |
| Mandarin (CN) | 87.2 |
| Catalan | 86.7 |
| Hungarian | 86.6 |
| Ukrainian | 86.4 |
| Greek | 86.3 |
| Bulgarian | 85.7 |
| Arabic | 84.9 |
| Serbian | 84.3 |
| Macedonian | 84.2 |
| Cantonese (HK) | 84.1 |
| Latvian | 83.3 |
| Slovenian | 83.2 |
| Hindi | 83.1 |
| Galician | 82.0 |
| Danish | 81.9 |
| Urdu | 79.6 |
| Slovak | 76.6 |
| Hebrew | 76.5 |
| Finnish | 75.4 |
| Azerbaijani | 75.2 |
| Lithuanian | 72.4 |
| Estonian | 70.1 |
| Nynorsk | 69.3 |
| Welsh | 68.0 |
| Punjabi | 65.3 |
| Afrikaans | 64.0 |
| Persian | 62.6 |
| Basque | 61.1 |
| Vietnamese | 60.1 |
| Bengali | 59.7 |
| Nepali | 59.6 |
| Marathi | 58.8 |
| Belarusian | 56.3 |
| Kazakh | 52.4 |
| Armenian | 51.9 |
| Swahili | 48.8 |
| Tamil | 48.5 |
| Albanian | 44.3 |
FLEURS — large-v3 (Expert Accuracy) (%)
| Language | Accuracy (%) |
|---|---|
| Spanish | 97.2 |
| Italian | 97.0 |
| Korean | 96.9 |
| Portuguese | 95.9 |
| English | 95.9 |
| Polish | 95.4 |
| Catalan | 95.2 |
| Japanese | 95.1 |
| German | 95.1 |
| Russian | 95.0 |
| Dutch | 94.8 |
| French | 94.7 |
| Indonesian | 93.9 |
| Ukrainian | 93.6 |
| Turkish | 93.3 |
| Malay | 92.7 |
| Swedish | 92.4 |
| Mandarin | 92.3 |
| Finnish | 92.3 |
| Norwegian | 92.2 |
| Romanian | 91.8 |
| Thai | 91.6 |
| Vietnamese | 91.4 |
| Slovak | 90.8 |
| Arabic | 90.4 |
| Czech | 89.9 |
| Croatian | 89.2 |
| Greek | 89.1 |
| Serbian | 88.4 |
| Danish | 88.0 |
| Bulgarian | 87.5 |
| Hungarian | 87.1 |
| Filipino | 87.0 |
| Bosnian | 87.0 |
| Galician | 86.9 |
| Macedonian | 85.3 |
| Hindi | 83.0 |
| Estonian | 81.9 |
| Slovenian | 81.7 |
| Tamil | 81.7 |
| Latvian | 80.6 |
| Azerbaijani | 80.3 |
| Urdu | 79.4 |
| Lithuanian | 76.3 |
| Hebrew | 73.9 |
| Welsh | 71.4 |
| Persian | 70.6 |
| Icelandic | 69.6 |
| Kazakh | 67.6 |
| Afrikaans | 67.6 |
| Kannada | 67.0 |
| Marathi | 65.9 |
| Swahili | 65.9 |
| Telugu | 60.7 |
| Maori | 60.2 |
| Nepali | 59.8 |
| Armenian | 57.8 |
| Belarusian | 57.5 |
| Gujarati | 52.6 |
| Punjabi | 51.5 |
| Bengali | 50.0 |
Source: WER/CER values visualized for Whisper large-v3 vs large-v2 on Common Voice 15 and FLEURS. Accuracy listed is 100 − reported error rate (WER or CER).
Have suggestions or feedback for EKHOS AI?
We’d love to hear from you! Please email us at feedback@ekhos.ai with your suggestions or ideas for improvement. Your input helps us refine our services and deliver the best experience possible.