We Built a Voice Design API That's 6x Cheaper Than ElevenLabs

The Problem with TTS Today

Most text-to-speech APIs fall into two camps:

Expensive but good: ElevenLabs, Azure Neural TTS - great quality but $0.15-0.30/min
Cheap but limited: edge-tts, Google Cloud TTS - free/cheap but no voice customization

What if you could generate any voice you wanted from a text description alone - at 1/6th the cost of ElevenLabs?

That's what we built with VoxCPM2 on PixelAPI.

What We Built

A self-hosted VoxCPM2 TTS API with two modes:

Voice Design: Generate speech using only a text description. No reference audio needed.

(A warm elderly man, gentle voice) Hello, welcome to our store.

Voice Cloning: Upload a reference audio clip and clone the voice for consistent brand voices.

Voice Design Quality - Real Benchmarks

We tested 8 different voice profiles using vision QC on spectrograms:

Voice	Score	Notes
Young Woman (gentle)	8/10	Clear formants, good dynamics
Child (curious)	8/10	Good high-end presence, no artifacts
Villain (dark)	7/10	Minor HF noise, good menace
Elderly Man (wise)	7/10	Slightly weak volume but clean

Speed on Real Hardware

Benchmarked on RTX 6000 Ada:

Audio Duration	Generation Time	RTF
10 seconds	~4 seconds	0.38
25 seconds	~9 seconds	0.38

RTF of 0.38 means it generates faster than real-time.

Pricing - The 2x Cheaper Rule

Provider	Voice Design	Voice Cloning
ElevenLabs	$0.30/min	$0.30/min + subscription
Play.ht	$0.036/min	$0.10/min
PixelAPI	$0.050/min	$0.100/min

6x cheaper than ElevenLabs for voice design, and we're self-hosted.

API Example - Python

import requests

resp = requests.post("https://api.pixelapi.dev/v1/tts/generate",
    headers={"Authorization": "Bearer YOUR_KEY"},
    data={"text": "(A warm elderly man, gentle) Hello everyone.", "language": "en"}
)
job = resp.json()
print(f"Job: {job['id']}")