The Claire Speech Engine (CSE)

The Claire Speech Engine is a production-grade, lightweight speech synthesis library built in Python.

Installation

pip install claire-speech-engine

To install from source for development:

git clone https://github.com/Yukariii04/Claire-Speech-Engine.git
cd Claire-Speech-Engine
pip install -e ".[dev]"

Quick Start

1. Setup Backend & Pre-download Models

CSE uses KittenTTS (CPU-optimized, ONNX Runtime) as its acoustic synthesis backend on Python 3.10 – 3.12.

# Setup KittenTTS dependencies and pre-download models for offline use
cse setup

2. Examples

You can instantly scaffold a runnable example into your current directory to test your setup:

cse example
python example_kittentts.py

3. Interactive CLI

# List available KittenTTS models (Nano, Micro, Mini)
cse models

# Interactively select your default model
cse model

# Interactively select your default voice
cse voice

4. Python API

from cse import SpeechEngine

engine = SpeechEngine()
engine.load_backend("kittentts")
# Loads your saved CLI preference, or falls back to backend default ('expr-voice-2-f' / 'Bella')
engine.load_voice()

speech = engine.speak("Synthesis is now extremely simple and lightweight.")
if speech.success:
    print(f"Audio saved to: {speech.audio_path}")

Architecture & Vision

Read these documents to understand the core philosophy and design of the engine:

• Project Vision Document (PVD-001)
• Architecture Overview

System Overview & Roadmap

The Claire Speech Engine (CSE) is currently at v1.0.5:

1. CSE (Framework & Orchestration): Runtime lifecycle, streaming controllers, voice registries, and user CLI configuration.
2. CPE Baseline (Performance Reasoning Pipeline): Initial rule-based passes (meaning -> intent -> planning) that infer basic communicative intent and delivery from punctuation/structure to construct the canonical PerformanceGraph.
3. Acoustic Synthesis: Powered by KittenTTS (ONNX) with zero PyTorch runtime overhead.

Future Development:

1. Full CPE (Claire Performance Engine): Deep semantic understanding, emotion reasoning, context-aware dialogue planning, and rich prosody control beyond basic punctuation heuristics.
2. CAM (Claire Acoustic Model): Custom in-house acoustic model designed to natively interpret PerformanceGraph representations.

License

This project is licensed under the MIT License. See the LICENSE file for details.

Development

Running Tests

pytest tests/

Running Golden Tests

pytest tests/golden/test_perf_golden.py

See docs/Benchmarks/README.md for full performance targets and reports.