REVERIE · TTS
Prerecordings
Maintain the cached audio for frequently requested text. These endpoints alter the prerecorded cache; each request requires an API key in the x-api-key header.
Incremental refresh
Rebuilds the cache for all active speakers using the top 100 texts from the latest two windows. Already cached texts are reused; only new texts are generated. The request returns after local files and the RAM cache are swapped. Versioned upload continues in the background. There is no request body.
curl -X POST "https://api.lucidweights.ai/v1/tts/prerecordings/refresh" \ -H "x-api-key: YOUR_API_KEY"
Full refresh
Regenerates every selected top text without reusing existing audio, so reference-audio updates are reflected in the cache. All body fields are optional. The response contains a message status string.
| Field | Type | Default | Description |
|---|---|---|---|
| speakers | string[] | all active | Restrict refresh to these speakers |
| top_n | integer | 100 | Number of top texts to regenerate |
| gen_concurrency | integer | 2 | Parallel generation workers |
curl -X POST "https://api.lucidweights.ai/v1/tts/prerecordings/refresh/manual" \
-H "x-api-key: YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{"speakers":["alice"],"top_n":50,"gen_concurrency":2}'Sync replicas
Reconciles the in-memory cache with stored manual entries and the latest automatic refresh snapshot. The versioned entry wins when a speaker and text collide. Only new or changed audio is downloaded; removed entries are dropped. No TTS engine is needed on the current pod, but the endpoint returns 503 if the speaker service is unavailable on this host. There is no request body.
curl -X POST "https://api.lucidweights.ai/v1/tts/prerecordings/sync" \ -H "x-api-key: YOUR_API_KEY"