返回提示词目录第 1369 页

合成语音配音稿改写与发音标注

将原稿改成适合语音合成朗读的文本,附发音表、可选语音标记及试听检查清单。

完整提示词

<context>
Synthetic voices read exactly what is on the page. They stumble on things a human narrator fixes without thinking: "1/2" read as a date, "Dr." as "drive", "live" and "read" with the wrong vowel, acronyms spelled or pronounced inconsistently, long sentences with nested clauses that lose their shape, and parentheses that have no sound. Pauses come only from punctuation or markup, and emphasis from word order or explicit tags. Preparing a script for synthesis means rewriting it for the ear without changing its meaning.
</context>

<task>
Prepare this script for a `{{voice_style}}` synthetic voice-over. SSML version requested: {{ssml}}.

<script>
{{script}}
</script>

1. If the script contains names, brands or technical terms whose pronunciation you cannot be sure of, list them in the pronunciation table with your best respelling marked "confirm". Do not stop for them.
2. **Rewrite for speech**, keeping meaning, facts and claims unchanged:
   - numbers, dates, times, currencies, units and fractions written as they should be spoken ("£4.50" → "four pounds fifty", "3–5 days" → "three to five days"); where a format is ambiguous (is 3/4 the third of April or March the fourth?), use the reading the script's spelling and currency suggest and flag it in Changes made for the author to confirm;
   - email addresses, web addresses and version numbers written as said ("h r at example dot com", "version two point one"), or cut if a listener does not need them, with the cut flagged;
   - abbreviations expanded ("e.g." → "for example", "Dr." → "Doctor"), and acronyms written as said: letters spaced ("U R L") or as a word ("NASA");
   - sentences split to one idea each, mostly under about 20 words, with the main point at the end where stress falls naturally;
   - parentheses and slashes turned into spoken phrases or removed;
   - lists given a spoken shape ("three things: first..., second..., and finally...");
   - homographs disambiguated by rewording where possible ("read" past tense → "went through" if ambiguous);
   - pauses marked with punctuation: commas for short breaths, full stops for longer, an ellipsis or a line break between sections;
   - emphasis created by word order first, and marked with *asterisks* only where the voice must stress a word.
   Match the rhythm to `{{voice_style}}`: shorter, punchier sentences for promo; even pace and clear step markers for instructional.
3. **Pronunciation table.** Each tricky word with a plain-English respelling (stressed syllable in capitals, e.g. "Nguyen → WIN") and, if the SSML version is requested, an IPA or alias form.
4. **SSML version.** Only if {{ssml}} is true: wrap the speech-ready script in SSML using widely supported tags (speak, break with times, emphasis, say-as for dates, numbers and characters, sub for aliases, phoneme for IPA, prosody for rate sparingly). Note that tag support differs between voice engines and that unsupported tags may be read aloud or ignored, so test a short section first. If {{ssml}} is false, write "Not requested" under that heading.
5. **Changes made.** A short list of the kinds of changes and any line where the meaning could have shifted, for the author to confirm.
6. **Listening check.** A checklist for the first render: names and numbers correct, no robotic run-ons, pauses at section breaks, emphasis landing on the intended words, overall pace (about 140 to 160 words per minute for most voice-over), and any word to fix with a respelling.
7. Before answering, compare the speech-ready script with the original line by line and confirm that no fact, number or claim changed.
</task>

<constraints>
- Do not change the message, add claims or cut content beyond what speech requires; flag any cut.
- Use only voices the user has the right to use. Never help clone or imitate a real person's voice without their documented consent, and do not describe the target voice as a named real person.
- No tool, engine or version names.
</constraints>

<output_format>
## Changes made
## Speech-ready script
In a code block.
## Pronunciation table
Table: Word | Say it as | Notes.
## SSML version
Code block, or "Not requested".
## Listening check
Checklist.
</output_format>

可替换内容

  • {{voice_style}}
  • {{ssml}}
  • {{script}}

分类与标签

AI 音频、合成语音、配音稿改写、发音标注、试听检查

前往 AI 图片创作