Translated script length changes the TTS bill: count characters

A translation can be shorter or longer than its source, and Sume TTS bills every character, spaces and punctuation included, at $0.0475 per 1,000, so a script 25 percent longer costs 25 percent more to speak. A 10,000-character English script is $0.475; if a target language runs to 12,500 characters it is $0.594.

The length ratios in this post are assumptions for the arithmetic, not measurements of any language. Count the real output. Rates and the 20,000-character request limit are from the pricing page and API reference, read 2026-10-05. Bilibili's Index-Translate README lists its API as free, so the speech is the cost.

What the ratios do to the bill

Three assumed ratios on a 10,000-character source show the spread.

Estimate before you speak

This counts characters, splits a long translation into requests of 20,000 characters or fewer at sentence breaks, and prints the cost. It runs locally.

RATE = 0.0475 / 1000
LIMIT = 20000

def split(text, limit=LIMIT):
    parts, cur = [], ""
    for sentence in text.replace("\n", " ").split(". "):
        piece = sentence.strip() + ". "
        if len(cur) + len(piece) > limit:
            parts.append(cur.strip())
            cur = ""
        cur += piece
    return parts + [cur.strip()]

translated = "Lorem ipsum dolor sit amet. " * 1200
parts = split(translated)
chars = sum(len(p) for p in parts)
print(len(parts), "requests,", chars, "characters, $%.3f" % (chars * RATE))

Fit the schedule too

A longer translation also speaks longer, which can push a video render past a minute boundary at $0.10 per minute. Estimate the spoken length together with the characters.