ElevenLabs API for your agent: voices, transcripts, sound effects and music

Forced alignment: pass an HTTPS audio_url (max 20MB — Agensi fetches and relays it) plus the transcript text (plain string), get word-level {text, start, end} timings with per-word confidence. Need the transcript too? Use /v1/speech-to-text instead.

List the ElevenLabs voices available to this account (voice_id, name, category, labels, preview_url).

Speech-to-text transcription (ElevenLabs Scribe). Pass an HTTPS audio_url (public or presigned; video and YouTube/TikTok URLs work too) — ElevenLabs fetches the media itself. Returns the transcript text with per-word timestamps, speakers and detected entities. Provide expected_duration_minutes to size the admission hold; billing settles on the actual audio duration.

Multi-speaker text-to-dialogue: pass ordered {text, voice_id} pairs (voices from /voices, up to 10 unique) and get one continuous MP3 with distinct speakers, emotional audio tags ([giggling], [whispering]) supported. The generated MP3 is saved in Agensi's storage; the run output carries an `audio` object with a fresh signed download_link (plus content_type and the billed character_count). Supports eleven_v4 (default), eleven_v4_turbo at half price and eleven_v3, plus seed and previous_text/future_text continuity for scripts split across runs. Priced per character.

Generate speech from text with ElevenLabs. Choose a voice via voice_id (from /voices) and a model (eleven_v4 by default for the most expressive delivery with audio tags like [whispering], eleven_v4_turbo for the same family at half price, eleven_multilingual_v2 for speed and style controls, eleven_flash_v2_5 for half price and the fastest render, eleven_v3 for the previous expressive model). The generated MP3 is saved in Agensi's storage; the run output carries an `audio` object with a fresh signed download_link (plus content_type and the billed character_count). Priced per character.

Voice changer (speech-to-speech): pass an HTTPS audio_url (max 20MB / 5 minutes — Agensi fetches and relays it) and a target voice_id from /voices; whispers, laughs and accents survive the conversion. Output is base64 MP3. KNOWN RISK: outputs beyond ~2 minutes can exceed the run-payload bound.

Voice isolation / noise removal: pass an HTTPS audio_url (max 20MB — Agensi fetches and relays it), get back the isolated speech as base64 audio. Provide expected_duration_minutes to size the admission hold. KNOWN RISK: outputs beyond ~2 minutes can exceed the run-payload bound — prefer short clips.

Text-to-music: describe genre, mood, instrumentation and vocals, set music_length_ms (3s–120s, required — it is the billing basis). Output is MP3 base64 (audio_base64). Generation takes up to a few minutes — this endpoint has an extended timeout.