Qwen3-TTS

Measured 2026-10-08. All 20 takes of the same text differ.

Fined:

• fatal · currency — dropped the minus sign (4 of 7 transcripts)

→ we sent “Your balance is -$1,204.50 as of today.”

→ it said “Your balance is nineus tweepounds and four thousand fifty dollars as of today.”

• fatal · large numbers — never said “eight point one billion” (6 of 7 transcripts)

→ we sent “Population reached 8,100,000,000 in 2023.”

→ it said “Population reached eight million one hundred million in twenty twenty-three.”

• fatal · units — never said “one and a half” (49 of 49 transcripts)

→ we sent “Add 1 1/2 cups of flour.”

→ it said “Add one one-half cups of flour.”

+ 27 more fined

Billed at $0.30 per audio hour. At the 14.7 characters per second measured on voice hudson, that is about $5.67 per 1M characters.