Аудио
Qwen3-TTS
Alibaba's open voice models: expressive, streaming-capable speech with free-form voice design.
★ 13,6 тыс. звёздApache-2.0
Смотреть руководство →
// факты о проекте
Факты о проекте
- Репозиторий
- remsky/Kokoro-FastAPI
- Звёзды GitHub
- 5 511
- Форки
- 908
- Лицензия
- Apache-2.0
- Последнее обновление на GitHub
- 1 октября 2026
- Категория
- Аудио
- Тип страницы
- Полное руководство по установке
- Официальный сайт
- Try the live demo
// как это работает
Как это работает
- Run one Docker command
Pull a prebuilt image with the model baked in — CPU, NVIDIA GPU, ARM, and AMD variants exist. The server starts on port 8880.
- Point your client at it
Anything written for OpenAI's /v1/audio/speech endpoint works unchanged — just change the base URL to localhost:8880.
- Pick voices and languages
Choose from built-in voices, blend several into a mix, or generate a custom voicepack with the clone tuner.
- Generate with extras
Request SSML for fine control, or ask for per-word timestamps to build read-along experiences and captions.
// возможности
Почему его любят
OpenAI-compatible API
A drop-in /v1/audio/speech endpoint — swap the URL in existing apps and keep your code.
Eight languages
English (US/GB), Spanish, French, Hindi, Italian, Japanese, Brazilian Portuguese, and Mandarin Chinese.
Voice mixing & cloning
Blend voices with weighted combinations or tune custom voicepacks from samples.
SSML + phonemes
Fine-grained pronunciation control, plus endpoints to generate phonemes from text and back.
Timestamped captions
Per-word or per-chunk captions for read-along UIs, subtitles, and karaoke-style apps.
Runs anywhere
Prebuilt images for CPU, NVIDIA, ARM, and AMD GPUs — plus Apple Silicon support when run directly.
Advertisement
// запустите сами
Запустите сами
- Step 1 — CPU (laptops, servers)
Start the CPU image; the model is baked in, so there is nothing else to download.
docker run -p 8880:8880 ghcr.io/remsky/kokoro-fastapi-cpu:latest - Step 2 — NVIDIA GPU
Use the GPU image instead for much faster synthesis.
docker run --gpus all -p 8880:8880 ghcr.io/remsky/kokoro-fastapi-gpu:latest - Step 3 — Speak
POST to the OpenAI-style endpoint, e.g. http://localhost:8880/v1/audio/speech, with your text and voice choice.
// сценарии использования
Что можно создать
- Narrate audiobooks from EPUBs and PDFs.
- Voiceovers for videos and explainers.
- A self-hosted TTS backend for your app or assistant.
- Accessibility readers and read-along learning tools.
Readers working through the Audio shelf also stop at TeroSubtitler, Free, cross-platform subtitle editor with AI transcription tools. SayIt opens source voice typing for Windows — a Wispr Flow / Superwhisper alternative. The Audio shelf also holds Parrot — Meeting recorder designed for your Mac with a live AI copilot.
Advertisement
// вопросы и ответы
Вопросы и ответы
Is it free?
Yes — Apache-2.0 licensed and self-hosted. You pay nothing per character, unlike cloud TTS APIs.
How good is the quality?
Kokoro-82M is famous for punching far above its 82-million-parameter weight, with remarkably natural prosody for its size.
Which languages are supported?
Eight: English (US and GB), Spanish, French, Hindi, Italian, Japanese, Brazilian Portuguese, and Mandarin Chinese.
Kokoro-FastAPI or Qwen3-TTS?
Kokoro-FastAPI is lighter, faster, and dead simple via Docker. Qwen3-TTS offers larger, more expressive Alibaba models with streaming — pick by your quality-vs-speed tradeoff.
Can I use it commercially?
Yes — the Apache-2.0 license permits commercial use.
// продолжить изучение
Похожие инструменты
Локальные LLM
Ollama
The simplest way to run open language models on your own machine.
★ 182,1 тыс. звёздMIT
Смотреть руководство →Локальные LLM
Unsloth
Run and fine-tune LLMs and diffusion models locally — faster training, smaller memory footprint.
★ 77,2 тыс. звёздApache-2.0
Смотреть руководство →Аудио
kokoro-tts
Turn books and documents into natural speech with the Kokoro TTS model.
★ 1,9 тыс. звёздMIT
Смотреть руководство →Аудио
TTS-WebUI
One Gradio + React WebUI for dozens of TTS, voice-cloning and music models — Bark to Kokoro.
★ 3,3 тыс. звёздMIT
Смотреть руководство →Аудио
GPA
GPA — one tiny model that does ASR, TTS and voice conversion (General Purpose Audio).
★ 3,1 тыс. звёздApache-2.0
Смотреть руководство →// альтернативы
Альтернативы: Аудио
Другие проекты из категории «Аудио» с наибольшим числом звёзд в каталоге LibreAI — реалистичный список альтернатив, если Kokoro-FastAPI вам не подходит.
- MoneyPrinterTurbo★ 128,3 тыс.
利用 AI 大模型和自动化工作流，根据主题或关键词一键生成高清短视频。Generate HD short videos from a topic or keyword with an automated AI workflow.
- Whisper★ 109,9 тыс.
Robust Speech Recognition via Large-Scale Weak Supervision
- GPT-SoVITS★ 62,3 тыс.
1 min voice data can also be used to train a good TTS model! (few shot voice cloning)
- Real-Time-Voice-Cloning★ 60,2 тыс.
Clone a voice from 5 seconds of audio and generate arbitrary speech in real time
- whisper.cpp★ 54,1 тыс.
Port of OpenAI's Whisper model in C/C++
// ещё в категории audio
Больше проектов: Аудио
Соседние проекты из категории «Аудио» в каталоге, ближайшие по числу звёзд к Kokoro-FastAPI — продолжайте изучать полку по обе стороны от него.
- YouDub-webui★ 5,6 тыс.
Open-source AI video localization and dubbing for YouTube/Bilibili: speech recognition, subtitle translation, voice cloning, audio mixing and rendering. 开源 A…
- fastrtc★ 4,6 тыс.
FastRTC turns any Python function into a real-time audio or video stream over WebRTC or WebSockets
- whisper-diarization★ 5,7 тыс.
Automatic Speech Recognition with Speaker Diarization based on OpenAI Whisper
- MOSS-TTS-Nano★ 4,5 тыс.
MOSS-TTS-Nano: a 0.1B open multilingual speech model that clones voices and streams in real time on CPU
- Piper★ 5,7 тыс.
Fast and local neural text-to-speech engine
- WhisperLive★ 4,3 тыс.
WhisperLive: a nearly-live OpenAI Whisper transcription server for files, mics, and streams
Получить Kokoro-FastAPI
Бесплатно и открыто — забирайте исходный код, ставьте звезду репозиторию и начинайте создавать.