docs(tts): Dutch/Swedish Pro voices + overview language table + Pro pool path (DS-1009, DS-1010) - #415
Conversation
Dutch and Swedish carried a "voice-count data pending" caveat on the Pro model card, but the live /lightning-v3.1-pro/get_voices response already returns dedicated voices for both. Confirmed on prod: Dutch (7): sanne, femke, fenna (F), daan, lars, bram, stijn (M) Swedish (7): astrid, ebba, saga, alva (F), erik, anton, nils (M) Changes: - Pro model-card language table: nl and sv voice counts flipped from `-` to `7`; the "pending" caveat sentence deleted. - Pro model-card European Languages voice catalog: Dutch and Swedish rows added between Italian and Portuguese; section header count updated 61 -> 75. - Voices & Languages page European accordion: same additions and header count update, keeping the two pages in sync. - get_voices OpenAPI response example: two new sample voices (sanne for Dutch, astrid for Swedish) so the API-ref example is not stuck on Indian-only voices.
|
crim doesn't review pull requests automatically here. Comment |
|
crim review |
|
crim is reviewing this pull request. Findings will be posted shortly. |
![crim-app[bot]](https://avatars.githubusercontent.com/u/64185252?s=60&v=4){kind=small}
There was a problem hiding this comment.
Assessment
Mostly consistent documentation update; one table was left out of sync with the new Dutch/Swedish voices.
What this PR does
Adds Dutch (7) and Swedish (7) Pro voices to the Lightning v3.1 Pro documentation: two example voices in the OpenAPI get_voices response, new rows in the European-language voice tables, and updated aggregate counts (61→75 voices, 10→12 languages). Verified counts: Dutch 3F+4M=7, Swedish 4F+3M=7, European total 61+14=75.
Findings
- Low:
fern/products/waves/pages/text-to-speech/get-voice-models-langs.mdx:309— the bottom "Supported languages" table (Pro voice counts per language) is missing Dutch and Swedish rows despite them now having 7 voices each.
Findings outside the diff
These concern lines GitHub won't allow inline comments on (unchanged or outside this PR's diff).
-
fern/products/waves/pages/text-to-speech/get-voice-models-langs.mdx:309— Low: The bottom "Supported languages" table is now out of sync with the rest of this PR.This table lists per-language Pro voice counts (e.g.
| German | de | - | 7 |at line 307), but Dutch (nl) and Swedish (sv) are absent even though they now have 7 trained Pro voices each. Every other surface in this PR was updated; this table still omits them, so a reader filtering by language here won't see the new codes. Add the two rows after Italian for consistency:| Dutch | `nl` | - | 7 | | Swedish | `sv` | - | 7 |
Good to mergeNumbers all add up (I counted every voice so you didn't have to), but one table still thinks Dutch and Swedish are on backorder. Slot in those two rows and this is a clean roll-out. |
…09, DS-1010) DS-1009: overview language table said German/French/Italian/Portuguese/ Russian/Polish are unavailable on standard Lightning v3.1 (they work, verified with live calls) and omitted Dutch and Swedish entirely. The whole 200-line raw-HTML `<table>` was also blowing up the .md the docs tell AI agents to fetch. DS-1010: /get_voices is pool-scoped but the guide always called the Standard path, so a caller who read only the guide never learned about the Pro path and never saw the 249-voice Pro pool. The hyphen-vs- underscore trap was only documented deep in the spec description. Changes: - overview.mdx: replaced the raw HTML table with a clean markdown one. Standard v3.1 now shows Yes on all 20 accepted codes (10 European + 10 Indic); Pro v3.1 shows Yes on all 31 (Standard + Greek/Finnish/ Norwegian + 8 Asian & Middle Eastern). Deleted the "British and American Pro voices speak English only" claim since live tags list ~21 languages on those voices. - get-voice-models-langs.mdx: the "Fetch the full catalog" section now calls both pool paths, notes the hyphen-vs-underscore split, and demos filtering by recommendedLanguages so callers can find the Dutch and Swedish dedicated voices. em-dash sweep clean; fern check 0 errors, 11 pre-existing warnings.
abhishekmishragithub
changed the title
1 participant
Summary
The Lightning v3.1 Pro model card carried a "voice-count data pending as new Pro voice trains for those roll out" caveat on Dutch and Swedish, and both languages showed `-` in the voice-count table. Both languages are already fully live on `GET /waves/v1/lightning-v3.1-pro/get_voices`, with 7 dedicated voices each.
Verified live
Ran against `api.smallest.ai` on 2026-09-05 with my API key:
Changes
Test plan
Em-dash sweep clean on the added lines. `fern check`: 0 errors, 11 warnings (all pre-existing).