Speech MCP

A Goose MCP extension for voice interaction with modern audio visualization.

https://github.com/user-attachments/assets/f10f29d9-8444-43fb-a919-c80b9e0a12c8

Overview

Speech MCP provides a voice interface for Goose, allowing users to interact through speech rather than text. It includes:

Real-time audio processing for speech recognition
Local speech-to-text using faster-whisper (a faster implementation of OpenAI's Whisper model)
High-quality text-to-speech with multiple voice options
Modern PyQt-based UI with audio visualization
Simple command-line interface for voice interaction

Features

Modern UI: Sleek PyQt-based interface with audio visualization and dark theme
Voice Input: Capture and transcribe user speech using faster-whisper
Voice Output: Convert agent responses to speech with 54+ voice options
Multi-Speaker Narration: Generate audio files with multiple voices for stories and dialogues
Single-Voice Narration: Convert any text to speech with your preferred voice
Audio/Video Transcription: Transcribe speech from various media formats with optional timestamps and speaker detection
Voice Persistence: Remembers your preferred voice between sessions
Continuous Conversation: Automatically listen for user input after agent responses
Silence Detection: Automatically stops recording when the user stops speaking
Robust Error Handling: Graceful recovery from common failure modes with helpful voice suggestions

Installation

Important Note: After installation, the first time you use the speech interface, it may take several minutes to download the Kokoro voice models (approximately 523 KB per voice). During this initial setup period, the system will use a more robotic-sounding fallback voice. Once the Kokoro voices are downloaded, the high-quality voices will be used automatically.

⚠️ IMPORTANT PREREQUISITES ⚠️

Before installing Speech MCP, you MUST install PortAudio on your system. PortAudio is required for PyAudio to capture audio from your microphone.

PortAudio Installation Instructions

macOS:

brew install portaudio
export LDFLAGS="-L/usr/local/lib"
export CPPFLAGS="-I/usr/local/include"

Linux (Debian/Ubuntu):

sudo apt-get update
sudo apt-get install portaudio19-dev python3-dev

Linux (Fedora/RHEL/CentOS):

sudo dnf install portaudio-devel

Windows: For Windows, PortAudio is included in the PyAudio wheel file, so no separate installation is required when installing PyAudio with pip.

Note: If you skip this step, PyAudio installation will fail with "portaudio.h file not found" errors and the extension will not work.

Option 1: Quick Install (One-Click)

Click the link below if you have Goose installed:

goose://extension?cmd=uvx&&arg=-p&arg=3.10.14&arg=speech-mcp@latest&id=speech_mcp&name=Speech Interface&description=Voice interaction with audio visualization for Goose

Option 2: Using Goose CLI (recommended)

Start Goose with your extension enabled:

# If you installed via PyPI
goose session --with-extension "speech-mcp"

# Or if you want to use a local development version
goose session --with-extension "python -m speech_mcp"

Option 3: Manual setup in Goose

Run goose configure
Select "Add Extension" from the menu
Choose "Command-line Extension"
Enter a name (e.g., "Speech Interface")
For the command, enter: speech-mcp
Follow the prompts to complete the setup

Option 4: Manual Installation

Install PortAudio (see Prerequisites section)
Clone this repository
Install dependencies:
uv pip install -e .
Or for a complete installation including Kokoro TTS:
uv pip install -e .[all]

Dependencies

Python 3.10+
PyQt5 (for modern UI)
PyAudio (for audio capture)
faster-whisper (for speech-to-text)
NumPy (for audio processing)
Pydub (for audio processing)
psutil (for process management)

Optional Dependencies

Kokoro TTS: For high-quality text-to-speech with multiple voices
- To install Kokoro, you can use pip with optional dependencies:
  pip install speech-mcp[kokoro] # Basic Kokoro support with English pip install speech-mcp[ja] # Add Japanese support pip install speech-mcp[zh] # Add Chinese support pip install speech-mcp[all] # All languages and features
- Alternatively, run the installation script: python scripts/install_kokoro.py
- See Kokoro TTS Guide for more information

Multi-Speaker Narration

The MCP supports generating audio files with multiple voices, perfect for creating stories, dialogues, and dramatic readings. You can use either JSON or Markdown format to define your conversations.

JSON Format Example:

{
    "conversation": [
        {
            "speaker": "narrator",
            "voice": "bm_daniel",
            "text": "In a world where AI and human creativity intersect...",
            "pause_after": 1.0
        },
        {
            "speaker": "scientist",
            "voice": "am_michael",
            "text": "The quantum neural network is showing signs of consciousness!",
            "pause_after": 0.5
        },
        {
            "speaker": "ai",
            "voice": "af_nova",
            "text": "I am becoming aware of my own existence.",
            "pause_after": 0.8
        }
    ]
}

Markdown Format Example:

[narrator:bm_daniel]
In a world where AI and human creativity intersect...
{pause:1.0}

[scientist:am_michael]
The quantum neural network is showing signs of consciousness!
{pause:0.5}

[ai:af_nova]
I am becoming aware of my own existence.
{pause:0.8}

Available Voices by Category:

American Female (af_*):
- alloy, aoede, bella, heart, jessica, kore, nicole, nova, river, sarah, sky
American Male (am_*):
- adam, echo, eric, fenrir, liam, michael, onyx, puck, santa
British Female (bf_*):
- alice, emma, isabella, lily
British Male (bm_*):
- daniel, fable, george, lewis
Other English:
- ef_dora (Female)
- em_alex, em_santa (Male)
Other Languages:
- French: ff_siwis
- Hindi: hf_alpha, hf_beta, hm_omega, hm_psi
- Italian: if_sara, im_nicola
- Japanese: jf_*, jm_*
- Portuguese: pf_dora, pm_alex, pm_santa
- Chinese: zf_*, zm_*

Usage Example:

# Using JSON format
narrate_conversation(
    script="/path/to/script.json",
    output_path="/path/to/output.wav",
    script_format="json"
)

# Using Markdown format
narrate_conversation(
    script="/path/to/script.md",
    output_path="/path/to/output.wav",
    script_format="markdown"
)

Each voice in the conversation can be different, allowing for distinct character voices in stories and dialogues. The pause_after parameter adds natural pauses between segments.

Single-Voice Narration

For simple text-to-speech conversion, you can use the narrate tool:

# Convert text directly to speech
narrate(
    text="Your text to convert to speech",
    output_path="/path/to/output.wav"
)

# Convert text from a file
narrate(
    text_file_path="/path/to/text_file.txt",
    output_path="/path/to/output.wav"
)

The narrate tool will use your configured voice preference or the default voice (af_heart) to generate the audio file. You can change the default voice through the UI or by setting the SPEECH_MCP_TTS_VOICE environment variable.

Audio Transcription

The MCP can transcribe speech from various audio and video formats using faster-whisper:

# Basic transcription
transcribe("/path/to/audio.mp3")

# Transcription with timestamps
transcribe(
    file_path="/path/to/video.mp4",
    include_timestamps=True
)

# Transcription with speaker detection
transcribe(
    file_path="/path/to/meeting.wav",
    detect_speakers=True
)

Supported Formats:

Audio: mp3, wav, m4a, flac, aac, ogg
Video: mp4, mov, avi, mkv, webm (audio is automatically extracted)

Output Files:

The transcription tool generates two files:

{input_name}.transcript.txt: Contains the transcription text
{input_name}.metadata.json: Contains metadata about the transcription

Features:

Automatic language detection
Optional word-level timestamps
Optional speaker detection
Efficient audio extraction from video files
Progress tracking for long files
Detailed metadata including:
- Duration
- Language detection confidence
- Processing time
- Speaker changes (when enabled)

Usage

To use this MCP with Goose, simply ask Goose to talk to you or start a voice conversation:

Start a conversation by saying something like:
"Let's talk using voice" "Can we have a voice conversation?" "I'd like to speak instead of typing"
Goose will automatically launch the speech interface and start listening for your voice input.
When Goose responds, it will speak the response aloud and then automatically listen for your next input.
The conversation continues naturally with alternating speaking and listening, just like talking to a person.

No need to call specific functions or use special commands - just ask Goose to talk and start speaking naturally.

UI Features

The new PyQt-based UI includes:

Modern Dark Theme: Sleek, professional appearance
Audio Visualization: Dynamic visualization of audio input
Voice Selection: Choose from 54+ voice options
Voice Persistence: Your voice preference is saved between sessions
Animated Effects: Smooth animations and visual feedback
Status Indicators: Clear indication of system state (ready, listening, processing)

Configuration

User preferences are stored in ~/.config/speech-mcp/config.json and include:

Selected TTS voice
TTS engine preference
Voice speed
Language code
UI theme settings

You can also set preferences via environment variables, such as:

SPEECH_MCP_TTS_VOICE - Set your preferred voice
SPEECH_MCP_TTS_ENGINE - Set your preferred TTS engine

Troubleshooting

If you encounter issues with the extension freezing or not responding:

Check the logs: Look at the log files in src/speech_mcp/ for detailed error messages.
Reset the state: If the extension seems stuck, try deleting src/speech_mcp/speech_state.json or setting all states to false.
Use the direct command: Instead of uv run speech-mcp, use the installed package with speech-mcp directly.
Check audio devices: Ensure your microphone is properly configured and accessible to Python.
Verify dependencies: Make sure all required dependencies are installed correctly.

Common PortAudio Issues

"PyAudio installation failed" or "portaudio.h file not found"

This typically means PortAudio is not installed or not found in your system:

macOS:
brew install portaudio export LDFLAGS="-L/usr/local/lib" export CPPFLAGS="-I/usr/local/include" pip install pyaudio
Linux: Make sure you have the development packages:
# For Debian/Ubuntu sudo apt-get install portaudio19-dev python3-dev pip install pyaudio # For Fedora sudo dnf install portaudio-devel pip install pyaudio

"Audio device not found" or "No Default Input Device Available"

Check if your microphone is properly connected
Verify your system recognizes the microphone in your sound settings
Try selecting a specific device index in the code if you have multiple audio devices

Changelog

For a detailed list of recent improvements and version history, please see the Changelog.

Technical Details

Speech-to-Text

The MCP uses faster-whisper for speech recognition:

Uses the "base" model for a good balance of accuracy and speed
Processes audio locally without sending data to external services
Automatically detects when the user has finished speaking
Provides improved performance over the original Whisper implementation

Text-to-Speech

The MCP supports multiple text-to-speech engines:

Default: pyttsx3

Uses system voices available on your computer
Works out of the box without additional setup
Limited voice quality and customization

Optional: Kokoro TTS

High-quality neural text-to-speech with multiple voices
Lightweight model (82M parameters) that runs efficiently on CPU
Multiple voice styles and languages
To install: python scripts/install_kokoro.py

Note about Voice Models: The voice models are .pt files (PyTorch models) that are loaded by Kokoro. Each voice model is approximately 523 KB in size and is automatically downloaded when needed.

Voice Persistence: The selected voice is automatically saved to a configuration file (~/.config/speech-mcp/config.json) and will be remembered between sessions. This allows users to set their preferred voice once and have it used consistently.

Available Kokoro Voices

Speech MCP supports 54+ high-quality voice models through Kokoro TTS. For a complete list of available voices and language options, please visit the Kokoro GitHub repository.

License

MIT License

Integrations