Voice and localization checklist
Speech design, plus STT and TTS across locales.
DesignerDeveloperQA
Voice UX
- Support barge-in
- Short lists: about three options
- Read numbers back in chunks and confirm
- Reprompt ladder: rephrase, example, alternative
- Offer keypad or a human early
- Write for the ear: short sentences, no markdown or URLs
- Aim for first audio within about a second; stream every stage
STT per locale
- Set the locale; test language identification
- Custom vocabulary or phrase hints for names and codes
- Test on real call audio with diverse accents
- Tune endpointing per language
- Normalize spoken forms; use confidence and alternatives
- Measure WER on names, numbers and intents
TTS per locale
- A chosen, reviewed voice per locale
- Normalize numbers, dates, currency and abbreviations
- SSML and lexicons for names and loanwords
- Complete messages with plural and agreement rules, no fragment gluing
- Check pace and pauses with native listeners
Locale details
- Use tags like pt-BR, es-MX, en-IN
- Decide formality per locale
- Dates, time, separators, currency, phone grouping
- Handle language switches: detect, confirm, keep context
- Segment dashboards and tests by locale