Voice Clones
Cloned from ~1 minute of Hebrew reference audio. Listen to the original voice (before) alongside the TTS output (after) from each provider.
Herman
Original Voice (Reference Audio)
TTS Output (Voice Clones)
Corn
Original Voice (Reference Audio)
TTS Output (Voice Clone)
Stock Voices
Pre-built Hebrew voices from various TTS providers. No voice cloning — these are the provider's default offerings.
Microsoft Edge TTS
Free, built into Microsoft Edge. Tested with Avri (male) and Hila (female) at normal and 70% speed.
Avri (Male) — Normal Speed
Avri (Male) — 70% Speed
Hila (Female) — Normal Speed
Hila (Female) — 70% Speed
ElevenLabs
v3 model with language_code: "he". Note: multilingual v2 produces unintelligible Hebrew — v3 with explicit Hebrew is required.
Google Gemini
Gemini 2.5 Flash Preview TTS via Google AI Studio.
Resemble AI
Stock Hebrew preset voice. Custom voice clones require nekudot (diacritics) for intelligible output.
Summary
| Provider | Approach | Hebrew Quality | Notes |
|---|---|---|---|
| MiniMax | Voice cloning (Replicate) | Best | Cloned from ~1 min Hebrew reference; natural sounding |
| Edge TTS | Stock voices | Good | Microsoft's free TTS; Avri and Hila voices |
| Gemini | Stock voices | Good | Google AI Studio, Gemini 2.5 Flash Preview TTS |
| ElevenLabs | Stock voices (v3) | Good | Requires language_code: "he" |
| Chatterbox | Voice cloning | Poor | Gibberish — Dicta ONNX nikud library unavailable on Replicate |
| Resemble AI | Stock + cloning | Poor | Needs nekudot (diacritics) for intelligible output |