Xiaoxiao is a sensible Mandarin default, but a default should not erase the catalogue. Microsoft documents multiple Simplified Chinese speakers and styles in its speech support list, and Google documents both established and newer Chinese models in its voice catalogue.
Match speaker to format
An assistant reply, a financial explainer, a children’s story and a long article do not need the same delivery. Teams should audition several voices on the exact format, at the intended speed, with the intended punctuation. Gender labels alone do not predict authority, warmth or fatigue.
Separate model choice from availability
The community edge-tts client can request Microsoft online voices without an Azure key. Azure offers the supported API route. Nkomo uses Edge first and Azure second while retaining the selected voice ID, so an outage does not unexpectedly replace the speaker. A Google voice provides model diversity when both Microsoft paths fail.
Expose choices safely
Nkomo’s capability registry now returns the released voices for each locale. Clients should populate a voice selector from that registry and submit the exact ID. The API rejects a voice that does not belong to the chosen locale, preventing a stale or cross-locale selection from silently changing pronunciation.
Revisit the comparison
Xiaoxiao currently leads Nkomo’s configured WaveNet-A route. The next fair comparison should include Google’s current premium Chinese models, multiple Microsoft voices, blind native review and long-form samples.
The right product position is choice with a reliable default: Xiaoxiao first today, a visible catalogue for audience-specific decisions, and evidence strong enough to change the order tomorrow.
Comments
No comments yet.