Skip to content

README.md

Latest commit

 

History

History
39 lines (30 loc) · 1.29 KB

README.md

File metadata and controls

39 lines (30 loc) · 1.29 KB

FireRedTTS Accelerated Service

Accelerated FireRedTTS1 with RedServing

Setup

  1. Download the FireRedTTS model:

    # Note! this weight is for non-commercial usage only
huggingface-cli download --resume-download --repo-type model FireRedTeam/FireRedTTS-1S --revision fireredtts1s_4_chat --local-dir ./tts_4_chat

  2. Run service:

    docker run -td --name ttsserver --security-opt seccomp:unconfined -v "$(pwd)/tts_4_chat/pretrained_models:/workspace/models/redtts" -p 8081:8081 crpi-byegxpnnsibfy3j1.cn-shanghai.personal.cr.aliyuncs.com/fireredchat/fireredtts1-server:latest bash /workspace/run.sh --llm --svc_config_path /workspace/svc.yaml --port 8081 --http_uri=/v1/audio/speech

API Usage

Endpoint: POST /v1/audio/speech

Example using curl:

curl https://localhost:8081/v1/audio/speech \
    -H "Content-Type: application/json" \
    -d '{
    "input": "哈喽，你好呀～",
    "voice": "f531",
    "response_format": "mp3"
    }' --output audio.mp3

Response:

<audio binary>

Notice

This TTS model weight is not under Apache2.0 (FireRedTeam/FireRedTTS-1S, revision fireredtts1s_4_chat) and the voice f531 included are for non-commercial usage only. You should also not use any f531 derivatives to train/distill your own model.