MULTILINGUAL VOICE SELECTION WORKSHEET
Burki free resource | Checked October 7, 2026
Original home: https://burki.dev/blog/multilingual-text-to-speech-voice-selection

This is a planning and review template. No voice, call or provider outcome has
been tested by this worksheet. Download is free; usage and voice work can cost
money. Use fictional information and recordings only under an agreed policy.

1. SELECT THE TASK AND FACT REVISION
Business: __________________  Task: __________________
Approved fact/script revision: __________________
Meaning that must survive in every language: __________________
Excluded promises/actions: __________________
Staff fallback and supported staff language: __________________

WORKED EXAMPLE (fictional studio, not a customer result)
Task: explain opening hours and enquiry process.
Fact card: open Monday-Friday, 9am-5pm. Staff review appointment requests.
Do not claim an appointment is available or confirmed from opening hours alone.
English script reviewer approves meaning first. Hindi reviewer prepares and
reviews the corresponding Hindi script. New language remains PENDING.

2. INVENTORY EACH LANGUAGE PATH
Copy these rows for each path:
Language and relevant variety: __________________
Desired provider locale: __________________
Provider / exact model ID: __________________
Voice display name / actual accessible voice ID: __________________
Per-voice languages/accents inventory and date checked: __________________
Requested language/locale appears in that inventory? __________________
Product exposes this model and required controls? __________________
Recognition model/language, checked separately: __________________
Business instructions and operational policy revision: __________________
Qualified reviewer: __________________
Voice access / commercial license / account funding checked: __________________
Status: PLANNED / SOURCE CHECKED / CONFIGURED / REQUEST ACCEPTED / REVIEWED

CURRENT BURKI BOUNDARIES TO RECHECK BEFORE USE
Conversation engine: Standard pipeline · advanced provider choices ->
Speech settings -> Advanced voice stack -> Voice provider.
Cartesia Voice model shows sonic-3 / sonic-3-2025-10-27 as of October7.
Do not infer sonic-3.6 from the word Latest. Cartesia marks the 2025-10-27
snapshot sunset October20,2026; review any existing configuration before then.
Voice tuning -> Language is a base-language choice. Locale/accent controls and
Cartesia Automatic language option are not exposed by this walkthrough.
GPT Live separate TTS is Announcement voice, not its conversation voice.

3. FIX THE CONFIGURATION FOR ONE PATH
Assistant draft/name: __________________
Conversation engine: __________________
Selected synthesis provider/model/voice/language: __________________
Recognition and reasoning configurations: __________________
Credentials or managed account arrangement reviewed: __________________
Platform/provider/telephone budget and owner: __________________
Save revision/time: __________________
Keep this fixed while reviewing the selected voice. Do not mix voice, language,
model and policy changes into one unexplained result.

4. REVIEW CASES (ALL UNTIL RUN = UNTESTED)
A. Short greeting.
B. Opening hours with a time and weekday.
C. Same approved fact in a longer response.
D. Caller asks whether opening hours guarantee an appointment.
E. Request beyond the approved language/service scope.
F. Caller corrects or restates the question.
G. Agreed staff fallback.
H. Mixed-language text only if this is a real required case; assess separately
   from one-language generation, rather than assuming universal code switching.

CASE RECORD (copy per case and language)
Case ID/date: __________________
Exact approved script / actual response text: __________________
Actual model / voice / language / locale if truly supported: __________________
Provider request receipt, if available: __________________
Audio actually reviewed? __________________
Intelligibility: ACCEPT / REVISE / UNTESTED
Voice suitability: ACCEPT / REVISE / UNTESTED
Business meaning preserved: YES / NO / UNTESTED
Unexpected accent/language/word delivery: __________________
Recognition or action problem, separately: __________________
Reviewer / reason: __________________
Next change, one at a time: __________________

5. GO / REVISE / HOLD
Approved task/model/voice/language path: __________________
Unreviewed language paths remain: __________________
Required fixes / owner / retest date: __________________
Expected staff outcome and fallback: __________________
Provider/model lifecycle review date: __________________

Sources and current setup: original article above. Model coverage, voice
coverage, settings, request acceptance and reviewed caller outcome are separate.

Voice selection: use Assistant voice. For Saved custom voice, expand
Custom voice identifier and enter the actual permitted Custom voice ID.