Models

KugelAudio offers models across two generations, each optimized for different use cases.

Model Overview

Model	Generation	Sample Rate	Max Input	Access
`kugel-1-turbo`	Kugel 1	24kHz	5000 chars	Public
`kugel-1`	Kugel 1	24kHz	5000 chars	Public
`kugel-2-turbo`	Kugel 2	24kHz	10000 chars	Early Access
`kugel-2`	Kugel 2	24kHz	10000 chars	Early Access

Kugel 2 models are in early access. They support IPA notation and custom pronunciation dictionaries. Contact us to get access.

Kugel 1 Turbo

Recommended for: Voice agents, real-time conversations, interactive applications

The Kugel 1 Turbo model is optimized for ultra-low latency, delivering excellent quality while maintaining ~39ms time-to-first-audio.

Specifications

Sample Rate: 24kHz
Max Input Length: 4096 characters
Time to First Audio: ~39ms (at inference)

When to Use

Voice agents and conversational AI
Real-time streaming from LLMs
Interactive applications requiring immediate feedback
High-volume production workloads

Example

Python
JavaScript
cURL

audio = client.tts.generate(
    text="Hello! I'm your AI assistant. How can I help you today?",
    model_id="kugel-1-turbo",
    cfg_scale=2.0,
)

const audio = await client.tts.generate({
  text: "Hello! I'm your AI assistant. How can I help you today?",
  modelId: 'kugel-1-turbo',
  cfgScale: 2.0,
});

curl -X POST https://api.kugelaudio.com/v1/tts/generate \
  -H "Authorization: Bearer $KUGELAUDIO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "text": "Hello! I'\''m your AI assistant. How can I help you today?",
    "model_id": "kugel-1-turbo",
    "cfg_scale": 2.0
  }' \
  --output output.pcm

Kugel 1

Recommended for: Audiobooks, podcasts, premium content, voice cloning

The Kugel 1 model is our premium offering. It produces the highest quality output with more natural prosody, emotion, and expressiveness.

Specifications

Sample Rate: 24kHz
Max Input Length: 8192 characters
Time to First Audio: ~77ms
Real-time Factor: < 0.3x (3x faster than real-time)

When to Use

Audiobook production
Podcast generation
Marketing and promotional content
Any application where quality is the top priority

Example

Python
JavaScript
cURL

audio = client.tts.generate(
    text="In a hole in the ground there lived a hobbit. Not a nasty, dirty, wet hole...",
    model_id="kugel-1",
    cfg_scale=2.5,  # Higher CFG for more expressive output
)

const audio = await client.tts.generate({
  text: "In a hole in the ground there lived a hobbit. Not a nasty, dirty, wet hole...",
  modelId: 'kugel-1',
  cfgScale: 2.5,  // Higher CFG for more expressive output
});

curl -X POST https://api.kugelaudio.com/v1/tts/generate \
  -H "Authorization: Bearer $KUGELAUDIO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "text": "In a hole in the ground there lived a hobbit. Not a nasty, dirty, wet hole...",
    "model_id": "kugel-1",
    "cfg_scale": 2.5
  }' \
  --output output.pcm

Generation Parameters

Both models support the following parameters:

Parameter	Type	Default	Description
`cfg_scale`	float	2.0	Classifier-free guidance scale (1.0-5.0). Higher values = more expressive
`max_new_tokens`	int	2048	Maximum tokens to generate
`sample_rate`	int	24000	Output sample rate in Hz
`voice_id`	int	null	Specific voice to use

CFG Scale Guide

The cfg_scale parameter controls how closely the model follows the voice characteristics:

1.0-1.5: More natural, relaxed delivery
2.0: Balanced (default)
2.5-3.0: More expressive, dynamic
3.5-5.0: Maximum expressiveness, best for dramatic content

Listing Models

Python
JavaScript
cURL

models = client.models.list()

for model in models:
    print(f"{model.id}: {model.name}")
    print(f"  Description: {model.description}")
    print(f"  Max Input: {model.max_input_length} characters")

const models = await client.models.list();

for (const model of models) {
  console.log(`${model.id}: ${model.name}`);
  console.log(`  Description: ${model.description}`);
  console.log(`  Max Input: ${model.maxInputLength} characters`);
}

curl https://api.kugelaudio.com/v1/models \
  -H "Authorization: Bearer $KUGELAUDIO_API_KEY"

Choosing the Right Model

Choose Kugel 1 Turbo when...

Latency is critical
Building real-time applications
Processing high volumes
Streaming from LLMs

Choose Kugel 1 when...

Quality is the top priority
Creating pre-recorded content
Voice cloning applications
Emotional/expressive content

Choose Kugel 2 Turbo when...

You need IPA notation support
Custom pronunciation dictionaries are required
Low latency with next-gen quality

Choose Kugel 2 when...

Maximum quality with next-gen architecture
Full IPA and dictionary support
Longer input texts (up to 10k chars)

Getting Started

Speech Generation

Voices

Integrations

Self-Hosted

SDK Reference

Model Overview

Kugel 1 Turbo

Specifications

When to Use

Example

Kugel 1

Specifications

When to Use

Example

Generation Parameters

CFG Scale Guide

Listing Models

Choosing the Right Model

Choose Kugel 1 Turbo when...

Choose Kugel 1 when...

Choose Kugel 2 Turbo when...

Choose Kugel 2 when...

Getting Started

Speech Generation

Voices

Integrations

Self-Hosted

SDK Reference

​Model Overview

​Kugel 1 Turbo

​Specifications

​When to Use

​Example

​Kugel 1

​Specifications

​When to Use

​Example

​Generation Parameters

​CFG Scale Guide

​Listing Models

​Choosing the Right Model

Choose Kugel 1 Turbo when...

Choose Kugel 1 when...

Choose Kugel 2 Turbo when...

Choose Kugel 2 when...

Model Overview

Kugel 1 Turbo

Specifications

When to Use

Example

Kugel 1

Specifications

When to Use

Example

Generation Parameters

CFG Scale Guide

Listing Models

Choosing the Right Model