mahimairaja/tts

This repository is a curated, awesome style list about text to speech, organized around one main split: real time streaming synthesis, used by voice agents that must start speaking within milliseconds, versus offline synthesis, used for audiobooks or dubs where quality matters more than speed. The list argues that most mistakes in building a text to speech product come from picking a tool built for the wrong side of that split, so every provider, model and benchmark in the list is labeled with which side it leans toward. The list covers the whole pipeline: how text gets normalized and converted to phonemes, how acoustic models predict the sound, and how vocoders turn that into an actual waveform. It also covers streaming and cancellation for live agents, voice cloning and voice design, controlling emotion and style, and support for multiple languages. Resources are tagged as beginner, intermediate, or advanced, and commercial sources are flagged so readers know when a guide has a sales angle. Explicitly left out of scope are speech to text, detecting when someone is speaking, detecting when a conversation turn ends, and the networking and telephony pieces around a voice pipeline, since the author considers those separate problems. Voice cloning and voice conversion are kept in scope, along with watermarking and the ethics of cloning a voice, since the author sees these as inseparable from actually shipping a synthetic voice today. The table of contents covers seventeen sections, including foundational concepts, commercial providers, open source models, streaming, cloning, multilingual support, the underlying architectures, key research papers, datasets, evaluation methods, audio codecs, watermarking, ethics and regulation, tutorials, tools, and community links. A resource only gets a spot in the list if it has been updated in the last year and is either vendor neutral or clearly marked when it comes from a company with something to sell.