public-mirrors/Kokoro-FastAPI
1
0
Fork 0
mirror of https://github.com/remsky/Kokoro-FastAPI.git synced 2025-04-13 09:39:17 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions
Dockerized FastAPI wrapper for Kokoro-82M text-to-speech model w/CPU ONNX and NVIDIA GPU PyTorch support, handling, and auto-stitching
8 commits 8 branches 13 tags 66 MiB
Python 74.9%
JavaScript 13.1%
CSS 7.2%
HTML 1.7%
Shell 1.2%
Other 1.9%
Find a file
remsky 0fb36bb1b2 fix: update benchmark results for processing time and output length
2024-12-30 06:16:55 -07:00
api/src feat: enabled support for stitching long outputs in TTS requests 2024-12-30 06:16:18 -07:00
examples fix: update benchmark results for processing time and output length 2024-12-30 06:16:55 -07:00
.gitignore Add initial implementation of Kokoro TTS API with Docker GPU support 2024-12-30 04:17:50 -07:00
docker-compose.yml Update Dockerfile and docker-compose.yml to add versions, specify Kokoro commit 2024-12-30 05:29:35 -07:00
Dockerfile fix: finalize pytorch version lock, git lfs 2024-12-30 06:09:17 -07:00
githubbanner.png Update README with performance benchmarks and usage examples; add benchmark plotting script 2024-12-30 04:53:29 -07:00
README.md Merge branch 'master' of https://github.com/remsky/Kokoro-FastAPI 2024-12-30 05:29:45 -07:00
requirements.txt Update Dockerfile and docker-compose.yml to add versions, specify Kokoro commit 2024-12-30 05:29:35 -07:00

README.md

Kokoro TTS Banner

Kokoro TTS API

Model Commit

FastAPI wrapper for Kokoro-82M text-to-speech model.

Dockerized with NVIDIA GPU support, simple queue handling via sqllite, and automatic chunking/stitching on lengthy input/outputs

Quick Start

# Start the API (will automatically clone source HF repo via git-lfs)
docker compose up --build

Test it out:

# From host terminal
python examples/test_tts.py "Hello world" --voice af_bella

Performance Benchmarks

Benchmarking was performed soley on generation via the API (no download) using various text lengths from 100 to ~10,000 characters, measuring processing time, token count, and output audio length. Tests were run on:

  • NVIDIA 4060Ti 16gb GPU @ CUDA 12.1
  • 11th Gen i7-11700 @ 2.5GHz
  • 64gb RAM
  • Randomized chunks from H.G. Wells - The Time Machine

Processing Time vs Output Length Processing Time vs Token Count

  • Average processing speed: ~3.4 seconds per minute of audio output
  • Efficient token processing: ~0.01 seconds per token
  • Scales well with longer texts, maintains consistent performance

API Endpoints

GET /tts/voices           # List available voices
POST /tts                 # Generate speech
GET /tts/{request_id}     # Check generation status
GET /tts/file/{request_id} # Download audio file

Example Usage

List available voices:

python examples/test_tts.py

Generate speech:

# Default voice
python examples/test_tts.py "Your text here"

# Specific voice
python examples/test_tts.py --voice af_bella "Your text here"

# Get file path without downloading
python examples/test_tts.py --no-download "Your text here"

Generated files are saved in:

  • With download: examples/output/
  • Without download: src/output/ (in API container)

Requirements

  • Docker
  • NVIDIA GPU + CUDA
  • nvidia-container-toolkit installed on host

Model

This API uses the Kokoro-82M model from HuggingFace.

Visit the model page for more details about training, architecture, and capabilities. I have no affiliation with any of their work, and produced this wrapper for ease of use and personal projects.

License

This project is licensed under the Apache License 2.0 - see below for details:

  • The Kokoro model weights are licensed under Apache 2.0 (see model page)
  • The FastAPI wrapper code in this repository is licensed under Apache 2.0 to match
  • The inference code adapted from StyleTTS2 is MIT licensed

The full Apache 2.0 license text can be found at: https://www.apache.org/licenses/LICENSE-2.0