Kokoro 82M is a lightweight, open-source text-to-speech model with only 82 million parameters that outperforms many larger TTS systems and paid cloud APIs. It runs locally on CPU (and fast on Apple Silicon), requires no GPU, supports 8 languages and 54 voices, and is licensed under Apache 2.0. Key advantages include low latency, offline operation, privacy, and near-zero cost at scale. Limitations include no native zero-shot voice cloning, neutral emotion output, and non-English voices still improving. A quick pip install and a sample Python script from the official repo are enough to get started.