Short version: we are not in closed beta yet not because we “missed a deadline”, but because we don’t want to ship a raw experience that disappoints users in the first minutes.
TTS is one of the core reasons. For HOLAS, this is not a side detail — it is at the center of the whole product experience.
Recently, we tested several options:
- Gemini
- OpenAI
- ElevenLabs
- Speechify
- the Belarusian model Bextts by TuteishyGPT
Here is the honest status today:
- Gemini — at our current stage, this is the best balance of cost + quality + speed. The trade-off: it is still a bit raw and sometimes does not finish phrases.
- OpenAI — very strong quality and fast generation, but the economics still do not work for us right now: several times more expensive than Google in our scenario.
- ElevenLabs — very clean pronunciation and genuinely impressive quality, but in our tests it is slower and also several times more expensive than Google.
- Speechify — at this point it does not provide the speed and pronunciation quality we need for our target level.
- Bextts / TuteishyGPT — for Belarusian pronunciation specifically, this is one of the most interesting and clean options. But there is infrastructure reality: to run it reliably, we need a powerful GPU server, which means hundreds of euros per month for a single server.
What’s next?
This is not a blocker, but a temporary stage: we continue building the full software layer around TTS so HOLAS is useful not only when voice is perfect, but in real day-to-day usage.
And yes — we are already considering entering closed beta even with reasonable, clearly communicated trade-offs.
Your feedback matters a lot to us:
- Would you use HOLAS if pronunciation issues happen from time to time?
- What monthly subscription price at launch would feel fair to you?
Share feedback in Instagram, LinkedIn, or Telegram — it really helps.