Choose a Voice for the Target Language
Kokoro Web currently focuses on English voices. Its English voice options should not be presented as Chinese, Japanese, or Spanish narration. For those languages, select a text-to-speech engine with an appropriate target-language voice; the script checks below help you prepare and review that engine's output. For an English version of a multilingual project, the Kokoro TTS multilingual workflow guide explains how to keep language versions organized.
Prosody via Punctuation and Segmentation
Prepare one idea at a time and listen for the transition between adjacent segments. A short sentence is a useful starting point, but a fixed duration such as twenty seconds is not a rule: sentence structure, speaking rate, and the selected model all affect where a natural boundary falls. Keep enough context around a clause so the voice does not sound as if it is starting over after every sentence.
Punctuation can suggest phrasing, but models do not all interpret commas, ellipses, or dashes the same way. Use the punctuation normal for the language, generate a small sample, and change one boundary at a time. Avoid adding repeated punctuation as a substitute for a pronunciation control.
Numbers, Dates, and Units
Write numbers the way you intend them to be spoken whenever a digit string has more than one plausible reading. A date like 03/04/2026 can mean different days in different locales; a decimal point, decimal comma, currency symbol, or abbreviation can also change how a voice reads a value. Prefer a locale-specific long form, then check it in the generated audio. Keep the written form consistent across headings, captions, and narration.
Loanwords and Names
Build a short pronunciation list before recording: people, place names, product names, acronyms, and words that change pronunciation between regions. Use a pronunciation dictionary if the engine supports one. If it does not, test a natural-language spelling or a phonetic note in a copy of the script, and keep the approved spelling in the final text where search, captions, or on-screen labels must remain correct. Do not assume that an improvised misspelling is an acceptable final form.
Language‑Specific Tips
Chinese
Keep technical terms consistent and split long sentences at clear clause boundaries. Decide how the script should read dates, decimals, percentages, and model names; digits and Latin abbreviations can sound different from their intended Mandarin reading. Have a fluent reviewer check names and tone-sensitive words in the final voice, not only in the text.
Japanese
Use punctuation to separate clauses and make topic shifts audible, while keeping particles attached to the phrase they belong to. Review kanji with multiple readings, counters, dates, and katakana loanwords in context. If the voice consistently chooses the wrong reading, check whether the engine offers a dictionary or reading annotation before rewriting visible text.
Spanish
Choose the intended regional variety when the project needs one, especially for vocabulary and pronunciation. Write ambiguous dates and decimal values in a clear spoken form, preserve accent marks, and decide whether an acronym should be read as a word or as separate letter names. Keep clauses direct so the sentence stress remains easy to hear.
Review the Audio with Native Context
Listen once for pronunciation and once for meaning. Check that the voice has not changed a number, dropped a short function word, or placed emphasis on the wrong part of a sentence. For customer-facing or instructional content, ask a fluent speaker from the intended region to review a representative sample and flag terms that sound technically correct but unnatural.
Maintain a small language sheet for each release: target locale, preferred names, acronym readings, date and measurement style, and approved translations. Reuse it when the script changes so different pages or episodes do not pronounce the same brand in different ways.
Evaluation Checklist
- Confirm the selected voice is designed for the target language and regional audience.
- Verify numbers, dates, acronyms, names, and product terms against an approved reference.
- Check transitions across segments and listen on the device where people will hear the narration.
- Ask a fluent reviewer to confirm comprehension and natural phrasing for important material.
Frequently Asked Questions
How should I handle numbers and dates?
Rewrite ambiguous dates, decimals, currencies, and measurements in the form a speaker should say, then listen to a sample in the target language.
What about loanwords and names?
Keep a pronunciation list, use pronunciation dictionaries when the selected engine supports them, and verify important names with a fluent reviewer.
How long should segments be?
There is no universal duration. Start with one sentence or a short paragraph, then adjust segment boundaries after listening to transitions in the selected voice.
Related Articles
Author: Kokoro Web Team • Last updated 2026-10-15