APIMaster ElevenLabs API
Use an APIMaster API key in the `Authorization: Bearer` header. These endpoints are under `/v1/elevenlabs` on APIMaster. The parameter references link to ElevenLabs.
Use an APIMaster API key in the Authorization: Bearer header. These endpoints are under /v1/elevenlabs on APIMaster. The parameter references link to ElevenLabs.
Upload a local file to the metered endpoints. Remote audio URLs and transcription webhooks are unavailable. Voice Design requires explicit preview text of at least 100 characters. Music and sound effects require a duration. Realtime conversation uses the APIMaster WebSocket, which handles upstream credentials. Send response.create after committing audio and wait for response.done.
Models and endpoints
| Capability | Model IDs | Upstream endpoint | Key parameters |
|---|---|---|---|
| Speech and dialogue | eleven_v4, eleven_v4_turbo, eleven_v3, eleven_v3_conversational, eleven_multilingual_v2, eleven_flash_v2_5, eleven_flash_v2, eleven_turbo_v2_5, eleven_turbo_v2 |
POST /v1/elevenlabs/text-to-speech/{voice_id} · POST /v1/elevenlabs/text-to-dialogue |
text · model_id · voice_settings · language_code · output_format · inputs[].voice_id · inputs[].text |
| File transcription | scribe_v3, scribe_v2, scribe_v2_medical |
POST /v1/elevenlabs/speech-to-text |
model_id · file · language_code · diarize · timestamps_granularity · keyterms · transcript_edit |
| Realtime transcription | scribe_v3_live, scribe_v2_realtime |
WSS /v1/elevenlabs/speech-to-text/realtime |
model_id · audio_format · commit_strategy · audio_base_64 · commit |
| Voice changer | eleven_multilingual_sts_v2, eleven_english_sts_v2 |
POST /v1/elevenlabs/voice-changer/{voice_id} |
audio · model_id · remove_background_noise · voice_settings · output_format |
| Voice design | eleven_ttv_v3, eleven_multilingual_ttv_v2 |
POST /v1/elevenlabs/text-to-voice/design |
model_id · voice_description · text · guidance_scale · seed |
| Sound effects | eleven_text_to_sound_v3, eleven_text_to_sound_v2 |
POST /v1/elevenlabs/sound-generation |
model_id · text · duration_seconds · loop · prompt_influence |
| Music | music_v2_5, music_v2, music_v1 |
POST /v1/elevenlabs/music |
model_id · prompt · composition_plan · music_length_ms · force_instrumental |
| Realtime conversation | eleven_realtime_v1_mini, eleven_realtime_v1_max |
WSS /v1/elevenlabs/realtime |
model · input_audio_buffer.append · response.create |
| Dubbing | dubbing_v2, dubbing_v1 |
POST /v1/elevenlabs/dubbing/project |
model_id · file · source_language · target_language · project_id |
Parameters and responses
- For speech,
voice_idselects the voice andmodel_idselects the synthesis model.language_codesets the language andvoice_settingsadjusts delivery. For dialogue, eachinputsentry supplies its own text and voice ID. - For file transcription, upload audio with
file. Usediarizefor speaker labels andtimestamps_granularityfor timestamp detail. For realtime transcription, send Base64 PCM chunks, commit a segment withcommit, and wait forcommitted_transcript. - Voice Changer preserves the original delivery and uses
voice_idfor the target voice. Voice Design returns three previews. Provide at least 100 characters intext. - For sound effects,
duration_secondsaccepts 0.5–30 seconds. For music, use eitherpromptorcomposition_plan. Withprompt, usemusic_length_msto set the duration andforce_instrumentalfor music without vocals. - Dubbing first returns a
project_id. Poll the project and language target and downloadoutputs.lossless_audioonce complete.
Examples
curl https://apimaster.ai/v1/elevenlabs/text-to-speech/JBFqnCBsd6RMkjVDRZzb \
-H "Authorization: Bearer $APIMASTER_API_KEY" \
-H 'Content-Type: application/json' \
-d '{"model_id":"eleven_v4","text":"Hello from ElevenLabs."}' \
--output speech.mp3
curl https://apimaster.ai/v1/elevenlabs/speech-to-text \
-H "Authorization: Bearer $APIMASTER_API_KEY" \
-F 'model_id=scribe_v3' -F 'file=@speech.mp3' \
-F 'diarize=true' -F 'timestamps_granularity=word'
curl https://apimaster.ai/v1/elevenlabs/music \
-H "Authorization: Bearer $APIMASTER_API_KEY" \
-H 'Content-Type: application/json' \
-d '{"model_id":"music_v2_5","prompt":"Soft piano jingle","music_length_ms":3000,"force_instrumental":true}' \
--output jingle.mp3
curl https://apimaster.ai/v1/elevenlabs/dubbing/project \
-H "Authorization: Bearer $APIMASTER_API_KEY" \
-F 'model_id=dubbing_v2' -F 'file=@speech.mp3' \
-F 'source_language=en' -F 'target_language=es'
curl "https://apimaster.ai/v1/elevenlabs/dubbing/project/$PROJECT_ID" \
-H "Authorization: Bearer $APIMASTER_API_KEY"
curl "https://apimaster.ai/v1/elevenlabs/dubbing/project/$PROJECT_ID/language/$LANGUAGE_ID" \
-H "Authorization: Bearer $APIMASTER_API_KEY"
Request formats and limits
Send synthesis, dialogue, voice design, sound effects, and music as application/json up to 4 MiB. Put output_format in the query string. Multi-speaker dialogue accepts eleven_v4 and eleven_v3. Append /stream to the synthesis, dialogue, voice changer, music, or audio isolation endpoint to read the response in chunks.
Uploads use multipart/form-data with one file. Use audio for voice changing and isolation, and file for transcription, alignment, stem separation, and dubbing. For raw mono 16 kHz 16-bit little-endian PCM, use file_format=pcm_s16le_16.
Music must be 3,000–600,000 milliseconds long. Supply music_length_ms with prompt. With composition_plan, set duration_ms on each segment and keep section durations fixed.
Realtime conversation example
Install websocket-client, set APIMASTER_API_KEY, and run this example. Send input after session.created. Play the audio as response.audio.delta or response.output_audio.delta events arrive. Read the turn status and usage from response.done.
import json
import os
from urllib.parse import urlsplit, urlunsplit
from websocket import create_connection
base = urlsplit("https://apimaster.ai/v1")
url = urlunsplit(("wss", base.netloc, base.path.rstrip("/") + "/elevenlabs/realtime",
"model=eleven_realtime_v1_mini", ""))
socket = create_connection(
url, header={"Authorization": "Bearer " + os.environ["APIMASTER_API_KEY"]},
timeout=30,
)
try:
assert json.loads(socket.recv())["type"] == "session.created"
socket.send(json.dumps({
"type": "conversation.item.create",
"item": {"type": "message", "role": "user",
"content": [{"type": "input_text", "text": "Say hello."}]},
}))
socket.send(json.dumps({"type": "response.create"}))
while True:
event = json.loads(socket.recv())
if event["type"] == "error":
raise RuntimeError(event["error"])
if event["type"] == "response.done":
print(event["response"]["status"], event["response"].get("usage"))
break
finally:
socket.close()
Standalone audio endpoints
POST /v1/elevenlabs/audio-isolation·audio,file_formatPOST /v1/elevenlabs/forced-alignment·file,textPOST /v1/elevenlabs/music/stem-separation·file,stem_variation_id(two_stems_v1,six_stems_v1)
Billing and compatibility
Speech is metered by characters. Transcription, voice conversion, dubbing, music and effects have separate duration-based units. Realtime conversations meter both connected minutes and tokens. Voice Design charges credits once for the preview text covering three samples. ElevenLabs plans to remove music_v1 on 2027-01-31 and eleven_text_to_sound_v2 on 2027-03-01.
