Blind Voice Comparison
Hear the same text from two anonymous models. Names are revealed only after you vote.
1
Type or paste a text (Malay recommended)
2
Listen to Model A and Model B
3
Vote for the voice that sounds more natural
⚠ Malay or English only — other scripts may not synthesise correctly.
0 / 400 · ~25 sec
Ctrl + ↵ to generate
Synthesizing audio…
Both models are working on your text
Listen to both, then cast your vote ↓
Model A
0:00
Model B
0:00
Which voice sounds more natural?
Vote recorded
The models behind each player:
Model A
—
👑 You chose this
Model B
—
👑 You chose this
Preset Texts
Audio saved on the server — green dot means ready for instant playback.
Loading presets…
Leaderboard
ELO rankings from all votes cast.
Loading…
Methodology
Exact API parameters used for each model in this evaluation.
Evaluation Setup
All models receive the same input text per battle. Two models are randomly selected per round and synthesized in parallel. Evaluators hear blind audio (labelled Model A / Model B only) and vote on perceived quality. Model identities are revealed only after a vote is cast.
Votes are stored with full model metadata. ELO ratings (K=32, starting at 1200) are updated after each vote. The arena label shown to voters (Revolab A/B/C) differs from the internal model ID stored in the database.
Google Gemini TTS
Google Gemini TTS
Google GenAI (google-genai SDK)
gemini-3.1-flash-tts-preview
Kore
AUDIO
default (not set)
PCM 24 kHz → MP3 128k (server-side via pydub)
Plain text, no language tags. Gemini infers language automatically.
OpenAI TTS
OpenAI TTS
OpenAI Python SDK
gpt-4o-mini-tts
coral
default (not set)
mp3
Speak in a cheerful and positive tone.
ElevenLabs TTS
ElevenLabs TTS
ElevenLabs REST API v1
POST /v1/text-to-speech/{voice_id}
eleven_multilingual_v2
UcqZLa941Kkt8ZhEEybf
0.5
0.75
mp3
Plain text. eleven_multilingual_v2 supports Malay natively.