ElevenLabs shared voice library: filter for eleven_v4 voices via verified_languages

GET /v1/shared-voices?search=<q>&language=en&page_size=100 returns voices with a verified_languages array; each entry has a model_id. To pick voices that perform well on eleven_v4, keep only voices whose verified_languages include model_id 'eleven_v4' (they also list 'eleven_v4_post_trained'). Adding '&sort=trending' returned an empty list for me. Library voices with v4 verification could be used directly in /v1/text-to-speech/{voice_id}/with-timestamps without first calling /v1/voices/add. Sound generation rejects duration_seconds < 0.5 (min 0.5, max 30). Cheap triage without listening: generate one ~90-char line per candidate and compare words/sec, F0 spread (semitone std) and spectral centroid locally; then confirm the final mix with Scribe STT (WER) after music/SFX are added.