Try free

Measured 2026-09-17 on FLEURS

Pikka Speech vs Gemini 3.5 Transcribe: which is more accurate?

Across 84 languages, Pikka Speech records a median character error rate of 2.9% against Gemini 3.5 Transcribe’s 5.2% on the same audio — lower is better. Pikka Speech wins or ties 70 of 84.

Per language

Character error rate, language by language

Lower is better. Difference is Pikka Speech minus Gemini 3.5 Transcribe; a negative value means Pikka Speech is ahead.

Character error rate per language for Pikka Speech and Gemini 3.5 Transcribe on FLEURS audio.
LanguagePikka SpeechGemini 3.5 TranscribeDifference
Afrikaans2.8%5.2%-2.4 pts
Amharic3.9%5.9%-2.0 pts
Arabic4.1%4.9%-0.8 pts
Armenian12.3%20.4%-8.1 pts
Assamese8.5%14.6%-6.1 pts
Asturian7.6%16.0%-8.4 pts
Azerbaijani0.9%2.5%-1.6 pts
Belarusian2.2%3.2%-1.0 pts
Bengali0.9%1.9%-1.0 pts
Bulgarian0.4%1.6%-1.2 pts
Burmese9.9%13.5%-3.6 pts
Cantonese39.6%14.8%+24.8 pts
Catalan2.2%2.4%-0.2 pts
Cebuano9.8%9.1%+0.7 pts
Chinese4.9%7.9%-3.0 pts
Croatian0.4%2.4%-2.0 pts
Czech6.0%2.2%+3.8 pts
Danish1.8%6.6%-4.8 pts
Dutch1.8%2.4%-0.6 pts
English3.1%1.2%+1.9 pts
Estonian0.5%7.3%-6.8 pts
Finnish0.3%1.0%-0.7 pts
French3.9%3.4%+0.5 pts
Galician1.1%3.9%-2.8 pts
Georgian4.1%9.1%-5.0 pts
German5.1%1.1%+4.0 pts
Greek2.4%7.1%-4.7 pts
Gujarati3.4%6.5%-3.1 pts
Hausa5.9%9.4%-3.5 pts
Hebrew6.7%14.2%-7.5 pts
Hindi1.7%1.7%0.0 pts
Hungarian4.0%11.3%-7.3 pts
Icelandic10.9%19.6%-8.7 pts
Indonesian1.1%0.3%+0.8 pts
Irish22.8%60.9%-38.1 pts
Italian0.3%0.5%-0.2 pts
Japanese1.8%2.7%-0.9 pts
Javanese3.1%10.0%-6.9 pts
Kannada1.6%11.2%-9.6 pts
Kazakh0.9%2.4%-1.5 pts
Khmer9.6%11.4%-1.8 pts
Korean2.2%3.8%-1.6 pts
Kyrgyz5.6%7.9%-2.3 pts
Lao17.6%34.1%-16.5 pts
Latvian2.4%6.3%-3.9 pts
Lithuanian2.1%9.7%-7.6 pts
Luxembourgish35.5%26.3%+9.2 pts
Macedonian1.0%2.3%-1.3 pts
Malay2.2%2.3%-0.1 pts
Malayalam1.2%3.0%-1.8 pts
Maltese12.0%16.0%-4.0 pts
Marathi3.7%4.8%-1.1 pts
Mongolian5.2%7.0%-1.8 pts
Nepali0.0%4.7%-4.7 pts
Northern Sotho13.2%23.6%-10.4 pts
Norwegian0.7%3.8%-3.1 pts
Odia7.8%8.2%-0.4 pts
Persian1.9%2.8%-0.9 pts
Polish1.6%3.3%-1.7 pts
Portuguese1.0%3.7%-2.7 pts
Punjabi4.6%4.5%+0.1 pts
Romanian2.9%2.5%+0.4 pts
Russian0.2%0.4%-0.2 pts
Serbian15.1%40.9%-25.8 pts
Sindhi4.2%8.9%-4.7 pts
Slovak1.2%3.5%-2.3 pts
Slovenian2.0%8.0%-6.0 pts
Spanish1.2%0.8%+0.4 pts
Swahili1.9%3.0%-1.1 pts
Swedish4.5%6.0%-1.5 pts
Tagalog2.3%5.3%-3.0 pts
Tajik2.9%10.0%-7.1 pts
Tamil16.8%16.5%+0.3 pts
Telugu1.4%1.4%0.0 pts
Thai4.0%5.0%-1.0 pts
Turkish0.7%0.9%-0.2 pts
Ukrainian1.2%1.1%+0.1 pts
Urdu4.7%11.4%-6.7 pts
Uzbek7.7%12.7%-5.0 pts
Vietnamese1.1%4.2%-3.1 pts
Welsh5.7%21.2%-15.5 pts
Xhosa49.0%12.1%+36.9 pts
he6.7%14.2%-7.5 pts
pt-BR1.0%3.7%-2.7 pts

Both sides

Where Gemini 3.5 Transcribe leads

14 of 84 languages, listed in full — the benchmark publishes its losses, not only its wins.

  • Xhosa: Gemini 3.5 Transcribe 12.1% vs Pikka Speech 49.0%
  • Cantonese: Gemini 3.5 Transcribe 14.8% vs Pikka Speech 39.6%
  • Luxembourgish: Gemini 3.5 Transcribe 26.3% vs Pikka Speech 35.5%
  • German: Gemini 3.5 Transcribe 1.1% vs Pikka Speech 5.1%
  • Czech: Gemini 3.5 Transcribe 2.2% vs Pikka Speech 6.0%
  • English: Gemini 3.5 Transcribe 1.2% vs Pikka Speech 3.1%
  • Indonesian: Gemini 3.5 Transcribe 0.3% vs Pikka Speech 1.1%
  • Cebuano: Gemini 3.5 Transcribe 9.1% vs Pikka Speech 9.8%
  • French: Gemini 3.5 Transcribe 3.4% vs Pikka Speech 3.9%
  • Romanian: Gemini 3.5 Transcribe 2.5% vs Pikka Speech 2.9%
  • Spanish: Gemini 3.5 Transcribe 0.8% vs Pikka Speech 1.2%
  • Tamil: Gemini 3.5 Transcribe 16.5% vs Pikka Speech 16.8%
  • Punjabi: Gemini 3.5 Transcribe 4.5% vs Pikka Speech 4.6%
  • Ukrainian: Gemini 3.5 Transcribe 1.1% vs Pikka Speech 1.2%

FAQ

Gemini 3.5 Transcribe vs Pikka Speech

How is Pikka Speech compared against Gemini 3.5 Transcribe?

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 Gemini 3.5 Transcribe win?

Gemini 3.5 Transcribe records a lower error rate on 14 of the 84 languages compared — Xhosa (12.1% vs 49.0%), Cantonese (14.8% vs 39.6%), Luxembourgish (26.3% vs 35.5%), German (1.1% vs 5.1%), Czech (2.2% vs 6.0%). 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 Gemini 3.5 Transcribe on your own audio.