LMS_Speaks

Text-to-Speech (TTS) plugin for LM Studio, built with the @lmstudio/sdk.

It exposes OpenAI-compatible TTS endpoints on a local HTTP server so that any LLM-environment client (e.g. Open WebUI, SillyTavern, Continue.dev, or your own scripts) can request speech synthesis without any third-party cloud service.

---

Features

Endpoint	Method	Description
/v1/audio/speech	POST	Convert text to audio (OpenAI-compatible)
/v1/audio/voices	GET	List available voices
/v1/models	GET	List available TTS models
/health	GET	Liveness probe

Supported TTS Engines

Engine	Value	Notes
System (OS built-in)	system	Uses say (macOS), PowerShell (Windows), or espeak-ng/espeak (Linux)
LM Studio TTS model	lmstudio	Planned – integrates with TTS models loaded in LM Studio

---

Requirements

• Node.js 18+
• LM Studio (latest release) when running as a plugin
• For the system engine:
  • macOS: built-in say command (no extra install needed)
  • Windows: built-in PowerShell + System.Speech (no extra install needed)
  • Linux: espeak-ng or espeak (sudo apt install espeak-ng)

---

Installation

# 1. Clone the repo
git clone https://github.com/ObtuseAglet/LMS_Speaks.git
cd LMS_Speaks

# 2. Install dependencies
npm install

# 3. Build
npm run build

---

Usage

Standalone mode (no LM Studio required)

# Default port 8880
STANDALONE=true npm start

# Custom port and voice
STANDALONE=true TTS_PORT=9000 TTS_VOICE=Alex npm start

As an LM Studio plugin

Step 1: Ensure LM Studio is running

1. Download and install LM Studio (latest version recommended).
2. Launch LM Studio and ensure it's running in the background.
3. Important: You do NOT need to load any language model to use this TTS plugin. The plugin works independently of loaded models.

Step 2: Install the plugin

1. Open LM Studio and navigate to the My Plugins section in the sidebar.
2. Click Add plugin from local folder (or + button).
3. Browse to and select the LMS_Speaks repository folder (where you cloned/extracted this project).
4. LM Studio will automatically detect the lms-manifest.json file and load the plugin.
5. You should see "lms-speaks" appear in your plugin list with a status indicator.

Step 3: Configure the plugin

1. Click on the lms-speaks plugin in the list to open its settings panel.
2. Configure the following options:
   • TTS Server Port: Port number for the HTTP server (default: 8880). Change this if another service is using port 8880.
   • Default Voice: The voice name to use when no voice is specified in requests (default: "default"). See Voice Configuration below.
   • TTS Engine: Currently only "System (OS built-in)" is available. Future versions will support LM Studio TTS models.
3. Click Apply or Save to activate the settings.

Step 4: Verify the plugin is running

Once configured, the plugin will start automatically. You can verify it's running by:

# Check the health endpoint
curl http://127.0.0.1:8880/health

# Should return: {"status":"ok"}

The TTS server is now ready to accept requests from any compatible client application!

Making requests

# Synthesize speech
curl -X POST http://127.0.0.1:8880/v1/audio/speech \
  -H "Content-Type: application/json" \
  -d '{"model":"tts-1","input":"Hello from LMS Speaks!","voice":"default"}' \
  --output speech.wav

# List voices
curl http://127.0.0.1:8880/v1/audio/voices

# List models
curl http://127.0.0.1:8880/v1/models

OpenAI-compatible client

import OpenAI from "openai";
import { writeFileSync } from "fs";

const client = new OpenAI({
  baseURL: "http://127.0.0.1:8880/v1",
  apiKey: "not-needed",
});

const speech = await client.audio.speech.create({
  model: "tts-1",
  voice: "default",
  input: "The quick brown fox jumps over the lazy dog.",
});

writeFileSync("output.wav", Buffer.from(await speech.arrayBuffer()));

---

Configuration

Environment variable	Default	Description
TTS_PORT	8880	Port the HTTP server listens on
TTS_VOICE	default	Default voice when none is specified in the request
STANDALONE	(unset)	Set to true to run without LM Studio

When running as an LM Studio plugin the port, voice, and engine can also be configured from within LM Studio's plugin settings UI.

---

Voice Configuration

Discovering available voices

To see all voices available on your system, query the voices endpoint:

curl http://127.0.0.1:8880/v1/audio/voices

Example response:

{
  "voices": [
    {
      "id": "Alex",
      "name": "Alex",
      "language": "en-US"
    },
    {
      "id": "Samantha",
      "name": "Samantha",
      "language": "en-US"
    }
  ]
}

Platform-specific voices

The available voices depend on your operating system:

macOS

macOS includes high-quality built-in voices. Common voices include:

• Alex - Male, English (US)
• Samantha - Female, English (US)
• Victoria - Female, English (US)
• Daniel - Male, English (UK)
• Karen - Female, English (Australia)

To see all available voices on macOS, run in Terminal:

say -v ?

To download additional voices:

1. Open System Settings > Accessibility > Spoken Content
2. Click System Voice dropdown
3. Select Manage Voices... and download additional language voices

Windows

Windows uses the Speech API with voices from the Microsoft Speech Platform:

• Microsoft David Desktop - Male, English (US)
• Microsoft Zira Desktop - Female, English (US)
• Microsoft Mark - Male, English (US)

To see available voices on Windows, run in PowerShell:

Add-Type -AssemblyName System.Speech
(New-Object System.Speech.Synthesis.SpeechSynthesizer).GetInstalledVoices() | ForEach-Object { $_.VoiceInfo.Name }

To add more voices:

1. Open Settings > Time & Language > Speech
2. Click Manage voices and download additional language packs

Linux

Linux uses espeak-ng (or espeak as fallback), which provides synthetic voices:

• default - System default voice
• Various language-specific voices (en, en-us, en-gb, es, fr, de, etc.)

To see available voices on Linux:

espeak-ng --voices

To install additional voices:

sudo apt install espeak-ng-data

Selecting a voice

There are two ways to select a voice:

1. Set a default voice globally

In LM Studio plugin mode: Configure the Default Voice field in the plugin settings.

In standalone mode: Use the TTS_VOICE environment variable:

STANDALONE=true TTS_VOICE=Samantha npm start

2. Specify voice per request

Include the voice parameter in your API request:

curl -X POST http://127.0.0.1:8880/v1/audio/speech \
  -H "Content-Type: application/json" \
  -d '{"model":"tts-1","input":"Hello!","voice":"Alex"}' \
  --output speech.wav

Fine-tuning speech output

You can control the speech characteristics using the speed parameter:

# Slower speech (0.5x speed)
curl -X POST http://127.0.0.1:8880/v1/audio/speech \
  -H "Content-Type: application/json" \
  -d '{"input":"Speaking slowly","voice":"Alex","speed":0.5}' \
  --output slow.wav

# Faster speech (1.5x speed)
curl -X POST http://127.0.0.1:8880/v1/audio/speech \
  -H "Content-Type: application/json" \
  -d '{"input":"Speaking quickly","voice":"Alex","speed":1.5}' \
  --output fast.wav

Speed range: 0.25 to 4.0 (where 1.0 is normal speed)

Note: The system TTS engine does not support pitch or other advanced tuning. For more control, consider using dedicated TTS models when the lmstudio engine becomes available in future versions.

---

Model Selection

This plugin exposes OpenAI-compatible models for broad compatibility with existing tools:

Available models

Model	Description
tts-1	Standard quality text-to-speech
tts-1-hd	High-definition quality (currently same as tts-1)

Note: Currently both models use the same underlying system TTS engine. The model parameter is accepted for OpenAI compatibility but does not affect output quality. Future versions will support actual model selection when using LM Studio TTS models.

Querying available models

curl http://127.0.0.1:8880/v1/models

Example response:

{
  "object": "list",
  "data": [
    {
      "id": "tts-1",
      "object": "model",
      "created": 1699000000,
      "owned_by": "system"
    },
    {
      "id": "tts-1-hd",
      "object": "model",
      "created": 1699000000,
      "owned_by": "system"
    }
  ]
}

Future: LM Studio TTS model support

In upcoming versions, you'll be able to:

1. Load TTS models directly in LM Studio (e.g., XTTS, Bark, or other local TTS models)
2. Select the lmstudio engine in plugin settings
3. Choose which loaded TTS model to use via the model parameter

This will enable higher quality, more customizable speech synthesis while keeping everything local.

---

Integration with TTS-Compatible Tools

LMS_Speaks provides an OpenAI-compatible API, making it easy to integrate with popular LLM interfaces that support text-to-speech.

Open WebUI

Open WebUI is a self-hosted ChatGPT-style interface.

Setup:

1. Ensure LMS_Speaks is running (verify with curl http://127.0.0.1:8880/health)
2. Open your Open WebUI instance
3. Navigate to Settings > Audio
4. Configure TTS settings:
   • TTS Engine: Select "OpenAI"
   • API Base URL: http://127.0.0.1:8880/v1
   • API Key: Leave empty or enter any dummy value (not validated)
   • TTS Model: tts-1 or tts-1-hd
   • TTS Voice: Choose from your available system voices (e.g., "Alex", "Samantha")
5. Click Save

Now when you receive messages in Open WebUI, you can click the speaker icon to hear them read aloud using your local TTS!

SillyTavern

SillyTavern is a popular UI for chatting with AI characters.

Setup:

1. Start LMS_Speaks (verify it's running)
2. Open SillyTavern
3. Click the Extensions menu (puzzle piece icon)
4. Navigate to TTS in the extensions panel
5. Configure the TTS provider:
   • Provider: Select "OpenAI"
   • API URL: http://127.0.0.1:8880/v1/audio/speech
   • Model: tts-1
   • Voice: Select your preferred voice from the dropdown
6. Enable Auto-play TTS if desired
7. Click Save Settings

Characters will now speak their responses using your local TTS system!

Continue.dev

Continue is an AI code assistant for VS Code and JetBrains IDEs.

Setup (VS Code):

1. Install the Continue extension from the VS Code marketplace
2. Open Continue settings (Ctrl/Cmd + Shift + P > "Continue: Open Config")
3. Add TTS configuration to your config.json:

{
  "tts": {
    "provider": "openai",
    "baseURL": "http://127.0.0.1:8880/v1",
    "apiKey": "not-needed",
    "model": "tts-1",
    "voice": "default"
  }
}

4. Save and reload Continue

Note: Continue's TTS support may vary by version. Check their documentation for the latest TTS configuration options.

Custom applications

Any application supporting OpenAI's TTS API can use LMS_Speaks:

Python example:

from openai import OpenAI

client = OpenAI(
    base_url="http://127.0.0.1:8880/v1",
    api_key="not-needed"  # API key not validated
)

response = client.audio.speech.create(
    model="tts-1",
    voice="Alex",
    input="Hello from Python!"
)

with open("output.wav", "wb") as f:
    f.write(response.content)

JavaScript/TypeScript example:

import OpenAI from "openai";
import { writeFileSync } from "fs";

const client = new OpenAI({
  baseURL: "http://127.0.0.1:8880/v1",
  apiKey: "not-needed",
});

const speech = await client.audio.speech.create({
  model: "tts-1",
  voice: "Alex",
  input: "Hello from JavaScript!",
});

writeFileSync("output.wav", Buffer.from(await speech.arrayBuffer()));

cURL examples

Direct API calls work from any system with curl:

# Basic synthesis
curl -X POST http://127.0.0.1:8880/v1/audio/speech \
  -H "Content-Type: application/json" \
  -d '{"model":"tts-1","input":"Hello world!","voice":"default"}' \
  --output hello.wav

# With custom speed
curl -X POST http://127.0.0.1:8880/v1/audio/speech \
  -H "Content-Type: application/json" \
  -d '{"model":"tts-1","input":"Speaking fast!","voice":"Alex","speed":1.5}' \
  --output fast.wav

---

Troubleshooting

Plugin not appearing in LM Studio

Problem: The plugin doesn't show up in the My Plugins section.

Solutions:

• Ensure lms-manifest.json exists in the repository root
• Verify you selected the correct folder (the one containing lms-manifest.json)
• Restart LM Studio and try adding the plugin again
• Check LM Studio logs for any error messages

Port already in use

Problem: Error message about port 8880 already being in use.

Solutions:

• Change the port in LM Studio plugin settings or use a different TTS_PORT value
• Find and stop the process using port 8880:

  # macOS/Linux
  lsof -ti:8880 | xargs kill -9
  
  # Windows (PowerShell)
  Stop-Process -Id (Get-NetTCPConnection -LocalPort 8880).OwningProcess

"No such voice" or voice not working

Problem: Specified voice is not recognized or produces no audio.

Solutions:

• Query available voices: curl http://127.0.0.1:8880/v1/audio/voices
• Use exact voice ID from the voices list
• Try using "default" as a fallback voice
• On macOS, verify the voice is installed in System Settings > Accessibility > Spoken Content
• On Windows, install additional voices from Settings > Time & Language > Speech
• On Linux, ensure espeak-ng or espeak is installed: sudo apt install espeak-ng

Connection refused or cannot reach server

Problem: Cannot connect to http://127.0.0.1:8880

Solutions:

• Verify the plugin/server is running: check LM Studio plugin status or console output
• Check the health endpoint: curl http://127.0.0.1:8880/health
• Ensure you're using the correct port (default 8880)
• Check firewall settings aren't blocking localhost connections
• Try http://localhost:8880 instead of 127.0.0.1

Audio quality issues

Problem: Generated speech sounds robotic or low quality.

Solutions:

• System TTS engines provide basic quality; this is expected behavior
• Try different voices (some are higher quality than others)
• Adjust the speed parameter (too fast or slow can degrade quality)
• For production use, consider waiting for LM Studio TTS model support in future versions

Plugin fails to start or crashes

Problem: Plugin shows error status or crashes on startup.

Solutions:

• Ensure Node.js 18+ is installed: node --version
• Rebuild the project: npm install && npm run build
• Check console logs for specific error messages
• Verify all dependencies installed correctly
• Try running in standalone mode first: STANDALONE=true npm start

Integration not working with client application

Problem: Open WebUI, SillyTavern, or other client can't use TTS.

Solutions:

• Verify LMS_Speaks is running and accessible
• Test the endpoint directly with curl first
• Check the client's TTS configuration matches the examples above
• Ensure you're using the correct base URL format (with /v1 for OpenAI compatibility)
• Some clients may require restart after configuration changes
• Check client logs for specific error messages

---

API Reference

POST /v1/audio/speech

Request body (JSON):

Field	Type	Required	Description
input	string	✅	Text to synthesize (max 4 096 chars)
model	string	❌	tts-1 or tts-1-hd (currently ignored)
voice	string	❌	Voice identifier; defaults to the configured default voice
response_format	string	❌	mp3, opus, aac, flac, wav, pcm; engine may return wav regardless
speed	number	❌	Speech speed multiplier 0.25–4.0 (default 1.0)

Response: raw audio bytes with an appropriate Content-Type header.

---

Development

# Run tests
npm test

# Watch-mode development
STANDALONE=true npm run dev

---

License

MIT