EfficientSpeech is an efficient, lightweight neural text-to-speech system designed for on-device inference, including resource-constrained devices such as the Raspberry Pi 4. It converts text to phonemes, predicts mel spectrograms with a shallow pyramid transformer architecture, and generates audio using a HiFi-GAN vocoder; it supports training, PyTorch inference, ONNX/TorchScript conversion, GPU acceleration, and LJSpeech preprocessing.

tts-audiobook-tool is a Python-based generative-AI audiobook creation application with a hotkey-driven console interface. It converts EPUB or text files into long-form audiobooks using numerous local or SGL-Omni-served TTS models, with voice cloning, segmentation, speech-to-text validation and retry logic, silence and loudness processing, chapter export, synchronized browser playback, realtime audio, LLM chat, and an optional REST server. It also includes a static JavaScript browser player and utilities for managing model-specific virtual environments.

Mix Space is a monorepo for an AI-powered, headless personal CMS stack supporting blogs, creator sites, and content websites. It includes a NestJS core server, React admin dashboard, native iOS client, APNs push relay, telemetry worker, and shared SDKs for APIs, databases, editing, AI, webhooks, CLI tooling, and migrations.

This repository documents a speaker-adaptation experiment for neural text-to-speech (TTS). It demonstrates fine-tuning a pretrained TTS model with approximately one minute of speech and ten minutes of training to synthesize voices resembling several celebrities, with links to audio samples and the underlying DCTTS implementation.

DurIAN is a PyTorch implementation of a Duration Informed Attention Network for text-to-speech synthesis. It provides a duration-aware encoder-decoder model and a baseline variant, training with phoneme-level duration alignments, mel-spectrogram reconstruction losses, TensorBoard logging, checkpointing, validation, and Griffin-Lim-based inference.

Resemble's Unity Plugin integrates the Resemble AI voice-generation API into Unity projects. It includes Unity package builds, editor preferences and help access, phoneme timing data, audio assets, and example scenes demonstrating phoneme-driven shaders and blendshape facial animation.

This repository provides a PyTorch implementation of a convolutional text-to-speech system based on the paper "Efficiently Trainable Text-to-Speech System Based on Deep Convolutional Networks with Guided Attention." It supports dataset downloading and audio preprocessing, training separate Text2Mel and SSRN spectrogram models, guided attention, checkpointing, TensorBoard logging, and speech synthesis for English LJ Speech and Mongolian Bible datasets.

NobodyWho is a Rust-based, on-device inference engine for running GGUF-format large language models locally and offline. It provides streaming chat, conversation and context management, grammar-based tool calling, embeddings, reranking, multimodal input, text-to-speech and speech-to-text, with bindings for Kotlin, Swift, Python, Flutter, React Native and Godot. Inference is powered by llama.cpp with Vulkan and Metal GPU acceleration.

ComfyUI-VoxCPM2 is a Python custom-node extension that integrates the VoxCPM2 tokenizer-free, diffusion autoregressive text-to-speech model into ComfyUI. It provides multilingual 48 kHz text-to-speech, voice design, controllable and transcript-based voice cloning, LoRA checkpoint loading, optional ASR transcription and denoising, model/VRAM management, and native LoRA training nodes.

UltraEval-Audio is an open-source unified evaluation framework for audio foundation models, covering both speech understanding and speech generation. It aggregates many benchmarks across speech, sound, medicine, and music, and provides benchmark management, built-in metrics, replication docs for popular models, and support for custom datasets and isolated runtime inference.