HeathenS_Talkings

Offline, modular Text-to-Speech with distinctive voice synthesis

A standalone Python toolkit for low-latency speech synthesis featuring voice cloning, distinctive voice character (e.g., Stephen Hawking's robotic prosody), and real-time streaming. Designed for local LLM integration, accessibility tools, and creative AI applications.

Features

• 🎙️ Voice Cloning — adapt TTS models to sound like specific speakers (5–10 min audio)
• ⚡ Low Latency — streaming synthesis for real-time conversation (~30ms/character)
• 📡 Streaming Output — SSE-compatible audio streaming for web/app integration
• 🔇 Offline Inference — 100% local, no cloud dependencies, no API keys
• 🎛️ Modular Design — use just TTS, just STT, or the full pipeline
• 🛠️ CLI + Python API — simple CLI commands or integrate via Python
• 🎯 Distinctive Voices — built for character synthesis (not just naturalness)

Quick Start

# Install
pip install heathen-tts

# Synthesize text to speech
hts tts "Hello, this is Stephen Hawking"

# Create a custom voice profile from local samples
hts clone hawking --samples ./samples/hawking_raw/

# Synthesize with a custom voice profile
hts clone-tts "Hello world" --profile models/voices/hawking.json -o hawking.wav

# Placeholder commands (not implemented yet)
# Transcribe audio to text
hts stt audio.wav

# Real-time listen & respond (with local LLM or manual input)
hts listen

# Start API server
hts serve --port 8000

See Getting Started for details.

Architecture

This project is built as 5 independent modules:

1. TTS Engine — convert text to audio (Glow-TTS, XTTS v2, Piper)
2. STT Engine — convert audio to text (Faster-Whisper, Wav2Vec)
3. Voice Cloning — fine-tune models or adapt speakers
4. Streaming — real-time audio output and pipelining
5. CLI & Integrations — user interface and optional bridges to larger systems

Each module is independently useful; together they form a complete I/O pipeline. See ARCHITECTURE.md for design details.

Use Cases

• 🤖 Personal Assistants — add a voice to local LLMs (works with Ollama, llama.cpp, vLLM)
• 🎬 Voice Synthesis — create character voices for media or storytelling
• ♿ Accessibility — generate audio for text in distinctive or preferred voices
• 🔬 Research — experiment with TTS model behavior and adaptation
• 🎮 Games & Creative Tools — real-time voice generation for NPCs or interactive media

Installation

# Basic TTS only
pip install heathen-tts[tts]

# TTS + STT
pip install heathen-tts[tts,stt]

# Full suite (TTS, STT, voice cloning, streaming)
pip install heathen-tts[full]

# Development
git clone https://github.com/IllTemperedMutatedSeaBass/HeathenS_Talkings
cd HeathenS_Talkings
pip install -e .[full]

Examples

See /examples for complete working examples:

• basic_tts.py — simple text-to-speech
• streaming_tts.py — real-time streaming audio
• voice_cloning.py — build/apply voice profiles
• http_client.py — call the local HTTP API

Documentation

• Getting Started
• Architecture
• Project Status
• Phase 1 Implementation Notes
• Publishing Guide

Integration with JesterAI

HeathenS_Talkings can optionally integrate as the C4 (TTS) component in JesterAI's distributed meeting intelligence system. This is not required — the project is fully standalone. Integration notes currently live in ARCHITECTURE.md.

Development

• Python 3.10+
• pyttsx3
• librosa, scipy, numpy
• Optional: Ollama (for local LLM text source)

Contributing: See CONTRIBUTING.md for workflow and guidelines.

License

Apache 2.0

Related Projects

• Glow-TTS — fast, deterministic TTS
• XTTS v2 — voice cloning and synthesis
• Faster-Whisper — fast speech-to-text
• Silero VAD — voice activity detection
• Piper TTS — lightweight TTS

Status: Version 0.1.0 published on PyPI. Current engine is the pyttsx3 baseline; STT/listen remain placeholder commands.