🎙️ Voice Design & TTS API

Generate natural speech from text using VoxCPM2 — a 2B parameter multilingual TTS model running on our self-hosted GPU infrastructure. Design any voice using text descriptions, or clone a voice from a reference audio sample. 30 languages, 48kHz output.

Quick Start

Voice Design (text → speech)

curl -X POST https://api.pixelapi.dev/v1/tts/generate \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -F "text=(A warm elderly man, gentle voice) Hello, welcome to our store." \
  -F "language=en"
# Response: { "id": "uuid", "status": "queued", "is_clone": false }

Voice Cloning (reference → speech)

curl -X POST https://api.pixelapi.dev/v1/tts/generate \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -F "text=This is my cloned voice speaking." \
  -F "[email protected]" \
  -F "language=en"
# Response: { "id": "uuid", "status": "queued", "is_clone": true }

Check Status

curl https://api.pixelapi.dev/v1/tts/status/YOUR_JOB_ID \
  -H "Authorization: Bearer YOUR_API_KEY"
# Response: { "id": "uuid", "status": "completed", "output_url": "https://..." }

Voice Description Guide

Use parentheses in the text field to describe the voice you want. VoxCPM2 generates speech matching your description.

Combine descriptions freely: (A wise elderly professor, warm and thoughtful)

API Endpoints

POST /v1/tts/generate

Generate speech from text. Returns a job ID — poll /v1/tts/status/{id} for completion.

GET /v1/tts/status/{generation_id}

Poll for generation status. When complete, returns output_url with the audio file.

GET /v1/tts/languages

List supported languages.

Supported Languages

VoxCPM2 supports 30 languages including: English, Hindi, Chinese, Spanish, French, German, Japanese, Korean, Russian, Arabic, Portuguese, Italian, Dutch, Polish, Turkish, Vietnamese, Thai, Indonesian, Malay, and more.

Pricing

Python SDK Example

import requests
import time

API_KEY = "your_api_key"
BASE = "https://api.pixelapi.dev"

# 1. Generate speech with Voice Design
resp = requests.post(f"{BASE}/v1/tts/generate",
    headers={"Authorization": f"Bearer {API_KEY}"},
    data={
        "text": "(A warm elderly man, gentle voice) Hello, welcome to our store.",
        "language": "en",
        "inference_timesteps": 10,
    }
)
job = resp.json()
print(f"Job: {job['id']}, credits: {job['credits_used']}")

# 2. Poll until complete
while True:
    status = requests.get(f"{BASE}/v1/tts/status/{job['id']}",
        headers={"Authorization": f"Bearer {API_KEY}"}
    ).json()
    print(f"Status: {status['status']}")
    if status["status"] == "completed":
        print(f"Audio URL: {status['output_url']}")
        break
    elif status["status"] == "failed":
        print(f"Error: {status.get('error')}")
        break
    time.sleep(2)

Voice Cloning Example

# Clone a voice from reference audio
resp = requests.post(f"{BASE}/v1/tts/generate",
    headers={"Authorization": f"Bearer {API_KEY}"},
    data={
        "text": "This is my cloned voice speaking.",
        "language": "en",
        "cfg_value": 2.0,  # cloning strength
    },
    files={"voice_ref": open("speaker.wav", "rb")}
)
job = resp.json()
print(f"Clone job: {job['id']}")

Comparison with Alternatives

Rate Limits

Voice Design and Voice Cloning share the same concurrent job limits as other tools. Pro plan users get higher concurrency. Generation time is typically 3–15 seconds depending on text length.