Text to Speech

Knox exposes ElevenLabs-compatible text-to-speech endpoints at https://api.knoxstudio.ai/v1 for KnoxStudio. Authenticate with your Knox API key (Authorization: Bearer sk-... or xi-api-key). The same routes are also served by https://api.knox.chat/v1.

This is not Realtime speech-to-speech, and it is not sending audio into /v1/chat/completions. TTS turns text into audio.

Typical flow:

1. List models — GET /v1/text-to-speech/models
2. Pick a voice_id from Voices (GET /v1/voices or GET /v2/voices)
3. Convert — POST /v1/text-to-speech/{voice_id}

https://api.knoxstudio.ai/v1
Authorization: Bearer sk-...

A commonly used public voice is 21m00Tcm4TlvDq8ikWAM (Rachel).

Models and billing

curl https://api.knoxstudio.ai/v1/text-to-speech/models

model_id	Latency / use	Max characters	Price
eleven_multilingual_v2	Default, 29 languages	10,000	$0.15 / 1K characters
eleven_v3	Most expressive, 70+ languages	5,000	$0.15 / 1K characters
eleven_v3_conversational	Voice agents, ~280ms	5,000	$0.10 / 1K characters
eleven_flash_v2_5	Ultra-low latency multilingual, ~75ms	40,000	$0.10 / 1K characters
eleven_flash_v2	Ultra-low latency English, ~75ms	30,000	$0.10 / 1K characters
eleven_turbo_v2_5	Deprecated; prefer Flash v2.5	40,000	$0.10 / 1K characters
eleven_turbo_v2	Deprecated; prefer Flash v2	30,000	$0.10 / 1K characters

Default model_id is eleven_multilingual_v2. Unauthenticated catalog requests are limited to 60/minute and may be cached for 5 minutes.

Create speech

text is required. Streaming variants do not support WAV output_format values.

cURL

curl -X POST "https://api.knoxstudio.ai/v1/text-to-speech/21m00Tcm4TlvDq8ikWAM?output_format=mp3_44100_128" \
     -H "Authorization: Bearer $KNOXCHAT_API_KEY" \
     -H "Content-Type: application/json" \
     -o speech.mp3 \
     -d '{
  "text": "Hello from Knox. This is text to speech.",
  "model_id": "eleven_multilingual_v2"
}'

Python

import os
import requests

voice_id = "21m00Tcm4TlvDq8ikWAM"
response = requests.post(
    f"https://api.knoxstudio.ai/v1/text-to-speech/{voice_id}",
    headers={
        "Authorization": f"Bearer {os.environ['KNOXCHAT_API_KEY']}",
        "Content-Type": "application/json",
    },
    params={"output_format": "mp3_44100_128"},
    json={
        "text": "Hello from Knox. This is text to speech.",
        "model_id": "eleven_multilingual_v2",
        "voice_settings": {
            "stability": 0.5,
            "similarity_boost": 0.75,
            "speed": 1.0,
        },
    },
)
response.raise_for_status()
open("speech.mp3", "wb").write(response.content)

Related endpoints

Goal	Endpoint
List / search voices	GET /v1/voices, GET /v2/voices
Stream audio	POST /v1/text-to-speech/{voice_id}/stream
Character timestamps	POST /v1/text-to-speech/{voice_id}/with-timestamps
Stream + timestamps	POST /v1/text-to-speech/{voice_id}/stream/with-timestamps
Chunked text over WebSocket	/v1/text-to-speech/{voice_id}/stream-input
Multiple concurrent contexts	/v1/text-to-speech/{voice_id}/multi-stream-input

Use HTTP convert when the full text is available. Use WebSocket when text arrives in chunks (for example an LLM stream). Plain HTTP to a WebSocket path returns 400.

Optional voice_settings: stability (default 0.5), similarity_boost (0.75), style (0), speed (1.0), use_speaker_boost (true). Up to 3 pronunciation_dictionary_locators and 3 previous_request_ids / next_request_ids. seed is 0–4294967295.

apply_text_normalization is auto, on, or off. language_code is ISO 639-1 and is ignored by eleven_multilingual_v2.

Errors

422 uses { "detail": [{ "loc": [...], "msg": "...", "type": "value_error" }] }. Other errors use { "detail": "..." }.

Common statuses: 401 missing/invalid key, 402 insufficient credits, 403 model not allowed, 422 missing text, over the character limit, or invalid output_format.