Voice similarity, supervision productivity, multilingual and data sovereignty — verified by benchmarks and real measurements, not claims.
OVERVIEW
VOICE SIMILARITY
On speaker similarity (SIM) — how close the cloned voice is to the original — we lead the market standard by a wide margin. It is hard to tell from a real person, so a rep can take over mid-call without any seam.
| Metric | Market standard | throw Voice |
|---|---|---|
| Speaker similarity (SIM · higher is better) | 61.3% | 85.4% +24.1%p |
| Word error rate (WER · lower is better) | 2.34% | 2.29% |
* English, Minimax-MLS benchmark vs a leading global voice AI. Korean is supported but not included in this benchmark test set.
CALL SUPERVISION
A person handles about 100 calls a day, one after another. With throw, AI runs the calls and one agent supervises 6 at once, lifting throughput 6×. In internal tests AI beat humans at every stage.
AI matched or beat humans at every stage of the call.
Follow-up and callback promises humans scored 0% on, AI still handles.
AI runs the calls while one agent supervises 6 concurrent at once.
※ 6 concurrent is the current capacity. Success rates are from an internal test of a specific outbound flow, human and AI under the same conditions.
MULTILINGUAL
Handle Korean, English and more at the same voice and quality. Our own training data produces natural pronunciation and intonation.
| Language | Market standard | throw Voice | Gap |
|---|---|---|---|
 English | 61.3% | 85.4% | +24.1%p |
 German | 56.8% | 79.2% | +22.4%p |
 French | 55.7% | 76.5% | +20.8%p |
 Cantonese | 67.0% | 83.5% | +16.5%p |
 Chinese | 67.7% | 81.4% | +13.7%p |
 Arabic | 70.6% | 79.1% | +8.5%p |
※ By voice similarity (SIM, higher is better), vs a leading global voice AI. throw Voice supports 30 languages and leads in every one. Korean is supported but not included in this benchmark test set.
SECURITY & DATA
The biggest risk in AI calls is call and customer data being sent to an outside AI server and leaking. throw Voice builds its own speech recognition, dialogue and synthesis and runs them itself, so security-strict insurance, finance and enterprises can adopt with confidence.
