Hey everyone, I’m currently testing a voice agent in Vapi using the Greek (EL) language, and I’m running into an issue with speech-to-text accuracy.
The agent barely understands anything I say and keeps asking me to repeat myself, even when I speak clearly. Here’s my current setup: * LLM: Anthropic Claude Sonnet 4 and also tested GPT-4o Cluster * Voice: ElevenLabs (Greek-supported voice) * Transcriber : Talkscriber Whisper and also tested the two Transcribers from OpenAI I’m mainly looking for input from people who have worked with non-English / less common languages: - Has anyone tested Greek (EL) with Vapi? - Any tips for improving transcription accuracy with low-resource or less supported languages? - Any best practices around audio quality, sampling rate, or configuration? - Any tips or workarounds that improved accuracy for you? Really appreciate any ideas or experience you can share, thanks!