Measured 2026-09-17 on FLEURS
Pikka Speech vs Google Chirp 3: which is more accurate?
Across 77 languages, Pikka Speech records a median character error rate of 2.8% against Google Chirp 3’s 12.1% on the same audio — lower is better. Pikka Speech wins or ties 71 of 77.
Per language
Character error rate, language by language
Lower is better. Difference is Pikka Speech minus Google Chirp 3; a negative value means Pikka Speech is ahead.
| Language | Pikka Speech | Google Chirp 3 | Difference |
|---|---|---|---|
| Afrikaans | 2.8% | 10.7% | -7.9 pts |
| Amharic | 3.9% | 12.3% | -8.4 pts |
| Arabic | 4.1% | 11.5% | -7.4 pts |
| Armenian | 12.3% | 21.1% | -8.8 pts |
| Assamese | 8.5% | 42.7% | -34.2 pts |
| Asturian | 7.6% | 10.8% | -3.2 pts |
| Azerbaijani | 0.9% | 28.0% | -27.1 pts |
| Bengali | 0.9% | 10.2% | -9.3 pts |
| Bulgarian | 0.4% | 8.2% | -7.8 pts |
| Burmese | 9.9% | 7.9% | +2.0 pts |
| Cantonese | 39.6% | 12.7% | +26.9 pts |
| Catalan | 2.2% | 3.6% | -1.4 pts |
| Chinese | 4.9% | 12.8% | -7.9 pts |
| Croatian | 0.4% | 2.7% | -2.3 pts |
| Czech | 6.0% | 9.1% | -3.1 pts |
| Danish | 1.8% | 2.2% | -0.4 pts |
| Dutch | 1.8% | 14.2% | -12.4 pts |
| English | 3.1% | 9.7% | -6.6 pts |
| Estonian | 0.5% | 9.2% | -8.7 pts |
| Finnish | 0.3% | 8.8% | -8.5 pts |
| French | 3.9% | 16.8% | -12.9 pts |
| Galician | 1.1% | 6.1% | -5.0 pts |
| Georgian | 4.1% | 4.5% | -0.4 pts |
| German | 5.1% | 5.7% | -0.6 pts |
| Greek | 2.4% | 9.6% | -7.2 pts |
| Gujarati | 3.4% | 15.6% | -12.2 pts |
| Hausa | 5.9% | 7.4% | -1.5 pts |
| Hebrew | 6.7% | 27.0% | -20.3 pts |
| Hindi | 1.7% | 17.3% | -15.6 pts |
| Hungarian | 4.0% | 12.1% | -8.1 pts |
| Icelandic | 10.9% | 12.1% | -1.2 pts |
| Indonesian | 1.1% | 14.5% | -13.4 pts |
| Italian | 0.3% | 0.8% | -0.5 pts |
| Japanese | 1.8% | 0.9% | +0.9 pts |
| Javanese | 3.1% | 6.0% | -2.9 pts |
| Kannada | 1.6% | 11.8% | -10.2 pts |
| Kazakh | 0.9% | 15.1% | -14.2 pts |
| Khmer | 9.6% | 14.3% | -4.7 pts |
| Korean | 2.2% | 8.3% | -6.1 pts |
| Kyrgyz | 5.6% | 31.0% | -25.4 pts |
| Lao | 17.6% | 10.1% | +7.5 pts |
| Latvian | 2.4% | 13.9% | -11.5 pts |
| Lithuanian | 2.1% | 17.9% | -15.8 pts |
| Luxembourgish | 35.5% | 36.1% | -0.6 pts |
| Macedonian | 1.0% | 9.3% | -8.3 pts |
| Malay | 2.2% | 22.2% | -20.0 pts |
| Malayalam | 1.2% | 11.8% | -10.6 pts |
| Maltese | 12.0% | 12.9% | -0.9 pts |
| Marathi | 3.7% | 26.3% | -22.6 pts |
| Mongolian | 5.2% | 5.9% | -0.7 pts |
| Nepali | 0.0% | 45.3% | -45.3 pts |
| Northern Sotho | 13.2% | 29.2% | -16.0 pts |
| Norwegian | 0.7% | 4.2% | -3.5 pts |
| Odia | 7.8% | 17.0% | -9.2 pts |
| Persian | 1.9% | 3.2% | -1.3 pts |
| Polish | 1.6% | 12.1% | -10.5 pts |
| Portuguese | 1.0% | 5.4% | -4.4 pts |
| Punjabi | 4.6% | 22.1% | -17.5 pts |
| Romanian | 2.9% | 60.1% | -57.2 pts |
| Russian | 0.2% | 3.2% | -3.0 pts |
| Serbian | 15.1% | 81.3% | -66.2 pts |
| Slovak | 1.2% | 5.0% | -3.8 pts |
| Slovenian | 2.0% | 23.5% | -21.5 pts |
| Spanish | 1.2% | 2.1% | -0.9 pts |
| Swahili | 1.9% | 7.8% | -5.9 pts |
| Swedish | 4.5% | 4.0% | +0.5 pts |
| Tagalog | 2.3% | 8.0% | -5.7 pts |
| Tamil | 16.8% | 28.8% | -12.0 pts |
| Telugu | 1.4% | 23.0% | -21.6 pts |
| Thai | 4.0% | 29.5% | -25.5 pts |
| Turkish | 0.7% | 44.9% | -44.2 pts |
| Ukrainian | 1.2% | 14.7% | -13.5 pts |
| Urdu | 4.7% | 29.2% | -24.5 pts |
| Uzbek | 7.7% | 14.1% | -6.4 pts |
| Vietnamese | 1.1% | 12.2% | -11.1 pts |
| Welsh | 5.7% | 25.8% | -20.1 pts |
| Xhosa | 49.0% | 44.8% | +4.2 pts |
Both sides
Where Google Chirp 3 leads
6 of 77 languages, listed in full — the benchmark publishes its losses, not only its wins.
- Cantonese: Google Chirp 3 12.7% vs Pikka Speech 39.6%
- Lao: Google Chirp 3 10.1% vs Pikka Speech 17.6%
- Xhosa: Google Chirp 3 44.8% vs Pikka Speech 49.0%
- Burmese: Google Chirp 3 7.9% vs Pikka Speech 9.9%
- Japanese: Google Chirp 3 0.9% vs Pikka Speech 1.8%
- Swedish: Google Chirp 3 4.0% vs Pikka Speech 4.5%
FAQ
Google Chirp 3 vs Pikka Speech
How is Pikka Speech compared against Google Chirp 3?
Both engines transcribed the same FLEURS audio — 10 utterances per language — through their own production interfaces, and results are scored as character error rate (CER) after normalising case and punctuation. Lower is better. Full per-language detail is on the speech accuracy page.
Where does Google Chirp 3 win?
Google Chirp 3 records a lower error rate on 6 of the 77 languages compared — Cantonese (12.7% vs 39.6%), Lao (10.1% vs 17.6%), Xhosa (44.8% vs 49.0%), Burmese (7.9% vs 9.9%), Japanese (0.9% vs 1.8%). Those rows are published in full on this page.
What does a difference in character error rate mean in practice?
CER is the share of characters an engine gets wrong: 3% means roughly three characters in every hundred differ from what was actually said. In live captions and interpretation, those errors concentrate in names, brands and technical terms — which is why the measured difference matters most on event audio.
Keep exploring
The full benchmark across all engines: speech recognition accuracy. Which languages we cover: language coverage. Or start a free test room.
Hear the difference on your own event
Create a free test room, speak a few sentences, and compare the live captions against Google Chirp 3 on your own audio.