High-fidelity neural text-to-speech engine with native Brazilian Portuguese voices and OpenAI API compatibility.
kokoro-82m-tts is an open source deployment packaging developed by Alex Santos. It transforms the raw Kokoro-82M library into an enterprise-ready Docker container featuring an OpenAI-compatible audio API, native pt-BR voice models, and automated test suites.
FastAPI gateway streaming neural audio chunks with CUDA acceleration and voice embedding caching.
Design choices, architectural patterns, and engineering decisions implemented for extreme reliability and developer experience.
Allows existing applications, agents, and pipelines built for OpenAI audio to switch to self-hosted Kokoro with zero code modifications.
Engineered specifically to overcome the lack of high-quality, open source Brazilian Portuguese neural speech voices.
Includes Dockerfile and compose templates optimizing NVIDIA GPU memory utilization for sub-second audio synthesis.
Validates voice selection, streaming playback, sample rates, and error responses before deployment.
Real-world usage, terminal configuration, and technical integration commands.
# cURL request to OpenAI-compatible endpoint
curl http://localhost:8880/v1/audio/speech -H "Content-Type: application/json" -d '{
"model": "kokoro",
"input": "Olá! Este é um teste de síntese de voz em português brasileiro.",
"voice": "pm_alex",
"response_format": "mp3"
}' --output speech.mp3Key clarifications regarding licensing, compatibility, deployment, and security.
Runs smoothly on NVIDIA GPUs with 4GB+ VRAM, or multi-core modern CPUs.
MP3, WAV, AAC, and Opus.
Yes, directly configure http://localhost:8880 as your TTS provider.
Yes, released under the Apache-2.0 open source license.