Kokoro TTS Engine Integration

Kokoro TTS is an ultra-fast, lightweight text-to-speech model with 82 million parameters. The model generates human-quality voice audio locally in under 100 milliseconds.

Local LLM Server Manager integrates Kokoro TTS on port 8880 and exposes an OpenAI-compatible speech proxy on port 5246.

---

Engine Overview

• Default Backend Port: :8880 (http://127.0.0.1:8880)
• Proxy Endpoint: POST http://127.0.0.1:5246/v1/audio/speech
• Model Footprint: ~350 MB (kokoro-v1.0.onnx + voices-v1.0.bin)
• Hardware Requirements: Runs efficiently on CPU or GPU with less than 1 GB VRAM
• Voice Catalog: 50+ built-in voices covering multiple accents and languages

---

Setting Up Kokoro TTS

You can install Kokoro TTS through the manager's modular component system:

1. Open the Local LLM Server Manager dashboard.
2. Select the Settings tab.
3. Locate the Modular Feature Packs section.
4. Find the Audio Feature Pack (ext_audio).
5. Click Install Audio Pack.
6. The manager executes the installation script (scripts/setup_kokoro_tts.ps1).
7. The installer sets up Python dependencies (kokoro-onnx, soundfile, fastapi, uvicorn) and downloads model weights (~350 MB).
8. Once installed, click Boot Audio Engine.
9. Verify the green Audio Engine status badge in the header.

NOTE

If you already run Kokoro-FastAPI independently, specify your custom executable path or Docker command in the Settings tab.

---

OpenAI-Compatible Speech Proxy

Local LLM Server Manager exposes a standard OpenAI /v1/audio/speech endpoint. Third-party applications can use local speech synthesis without code modifications.

API Endpoint Specification

• URL: http://localhost:5246/v1/audio/speech
• Method: POST
• Headers: Content-Type: application/json

Sample JSON Request Body

{
  "model": "kokoro",
  "input": "Local LLM Server Manager provides fast, private speech synthesis.",
  "voice": "af_heart",
  "response_format": "mp3",
  "speed": 1.0
}

Supported Parameters

Parameter	Type	Required	Description	Default
model	string	Yes	Model identifier (kokoro or tts-1).	kokoro
input	string	Yes	Text string to synthesize into speech.	N/A
voice	string	Yes	Target voice identifier (see voice catalog).	af_heart
response_format	string	No	Output audio container (mp3, wav, opus, flac).	mp3
speed	number	No	Speaking speed multiplier (0.5 to 2.0).	1.0

Testing with cURL

Open your terminal and run this command to generate speech:

curl -X POST http://localhost:5246/v1/audio/speech \
  -H "Content-Type: application/json" \
  -d "{\"model\":\"kokoro\",\"input\":\"Welcome to local speech synthesis!\",\"voice\":\"af_heart\"}" \
  --output welcome.mp3

TIP

Configure Open WebUI or LibreChat to use http://localhost:5246/v1 as their OpenAI TTS endpoint. Your chat assistants will speak responses aloud with zero cloud latency.

---

Voice Catalog (50+ Voices)

Kokoro uses standard naming prefixes to organize voice traits:

• First character (Language/Accent): a = American English, b = British English, j = Japanese, z = Mandarin Chinese, e = Spanish, f = French, h = Hindi, i = Italian.
• Second character (Gender): f = Female, m = Male.

Recommended English Voices

Voice ID	Accent	Gender	Description	Quality Rank
af_heart	American	Female	Warm, highly expressive flagship voice	Top Pick
af_bella	American	Female	Clear, energetic, professional	High
af_nicole	American	Female	Calm, relaxed conversational tone	High
af_sky	American	Female	Bright, modern assistant voice	High
am_adam	American	Male	Deep, clear narration voice	Top Pick
am_echo	American	Male	Balanced, smooth instructional voice	High
am_fenrir	American	Male	Resonant, authoritative tone	High
am_michael	American	Male	Natural conversational pacing	High
bf_alice	British	Female	Formal, articulate presentation tone	Top Pick
bf_emma	British	Female	Gentle, friendly British accent	High
bm_daniel	British	Male	Crisp, documentary narration style	Top Pick
bm_george	British	Male	Warm, distinguished tone	High

International Voices

Kokoro also includes multilingual voices:

• Japanese: jf_alpha, jf_gongitsune, jm_kumo
• Mandarin Chinese: zf_xiaobei, zf_xiaoni, zm_yunjian
• Spanish: ef_dora, em_alex, em_santa
• French: ff_siwis, fm_luc
• Italian: if_sara, im_nicola
• Hindi: hf_alpha, hm_omega

IMPORTANT

The af_heart voice offers the highest expressive fidelity. Select af_heart as your default voice for conversational agents.

---

Using Kokoro in Audio Studio

1. Open the Studio tab in the dashboard.
2. Select the Audio & Speech sub-tab.
3. Type or paste text into the Speech Input box.
4. Select your voice from the Voice Catalog dropdown.
5. Adjust the Speaking Speed slider.
6. Click Generate Speech.
7. Listen to the generated audio on the interactive waveform player.
8. Click Download Audio to save the file.

---

Related Documentation

• Engines & VRAM Orchestrator Overview
• Multimodal Audio & Music Studio Guide