steep imp
Well,
@elevenlabsio
is not reliable for many languages just yet, when using the multilingual v2 model.
Test
- Test 10 phrases in 6 languages
- Feed text in #XILabs API for genAI audio output
- Transcribe audio output via OpenAI's #Whisper
- Assert equality between input phrase and output phrase (they must be equal)
Here are my success rates :
🇺🇸 100%
🇪🇸 100%
🇩🇪 70%
🇫🇷 80%
🇯🇵 10%
🇨🇳 10%
Thus, seriously mulling on limiting my chatbot service to 🇺🇸🇪🇸.