Skip to content

Latest commit

 

History

History
258 lines (205 loc) · 12.4 KB

api.md

File metadata and controls

258 lines (205 loc) · 12.4 KB
Raw

API

使用 launch.py 脚本启动 api 服务之后，你可以在 http://localhost:7870/docs 下查看和简单测试 api

usage: launch.py [-h] [--cors_origin CORS_ORIGIN] [--no_playground] [--no_docs] [--exclude EXCLUDE]
                 [--compile] [--flash_attn] [--no_half] [--off_tqdm] [--device_id DEVICE_ID]
                 [--use_cpu {all,chattts,enhancer,trainer} [{all,chattts,enhancer,trainer} ...]]
                 [--lru_size LRU_SIZE] [--debug_generate] [--preload_models] [--host HOST] [--port PORT]
                 [--reload] [--workers WORKERS] [--log_level LOG_LEVEL] [--access_log] [--proxy_headers]
                 [--timeout_keep_alive TIMEOUT_KEEP_ALIVE]
                 [--timeout_graceful_shutdown TIMEOUT_GRACEFUL_SHUTDOWN] [--ssl_keyfile SSL_KEYFILE]
                 [--ssl_certfile SSL_CERTFILE] [--ssl_keyfile_password SSL_KEYFILE_PASSWORD]

Start the FastAPI server with command line arguments

options:
  -h, --help            show this help message and exit
  --cors_origin CORS_ORIGIN
                        Allowed CORS origins. Use '*' to allow all origins.
  --no_playground       Disable the playground entry
  --no_docs             Disable the documentation entry
  --exclude EXCLUDE     Exclude the specified API from the server
  --compile             Enable model compile
  --flash_attn          Enable flash attention
  --no_half             Disalbe half precision for model inference
  --off_tqdm            Disable tqdm progress bar
  --device_id DEVICE_ID
                        Select the default CUDA device to use (export CUDA_VISIBLE_DEVICES=0,1,etc might
                        be needed before)
  --use_cpu {all,chattts,enhancer,trainer} [{all,chattts,enhancer,trainer} ...]
                        use CPU as torch device for specified modules
  --lru_size LRU_SIZE   Set the size of the request cache pool, set it to 0 will disable lru_cache
  --debug_generate      Enable debug mode for audio generation
  --preload_models      Preload all models at startup
  --host HOST           Host to run the server on
  --port PORT           Port to run the server on
  --reload              Enable auto-reload for development
  --workers WORKERS     Number of worker processes
  --log_level LOG_LEVEL
                        Log level
  --access_log          Enable access log
  --proxy_headers       Enable proxy headers
  --timeout_keep_alive TIMEOUT_KEEP_ALIVE
                        Keep-alive timeout duration
  --timeout_graceful_shutdown TIMEOUT_GRACEFUL_SHUTDOWN
                        Graceful shutdown timeout duration
  --ssl_keyfile SSL_KEYFILE
                        SSL key file path
  --ssl_certfile SSL_CERTFILE
                        SSL certificate file path
  --ssl_keyfile_password SSL_KEYFILE_PASSWORD
                        SSL key file password

tips:

  • 所有参数均可在 .env.api 中以大写形式配置 （比如 no_docs => NO_DOCS）
  • 在命令行之后的参数优先级高于 .env 参数
  • launch.py 脚本启动成功后，你可以在 /docs 下检查 api 是否开启。

下面特殊接口的介绍

OpenAI API: v1/audio/speech

1. 语音合成 API

Endpoint

POST /v1/audio/speech

请求体

参数名 类型 默认值 描述
input string 必填 需要合成的文本
model string "chattts-4w" 模型名称
voice string "female2" 语音类型
response_format string "mp3" 响应音频格式，可选值：mp3, wav, ogg
speed float 1 音频播放速度，范围是 0.1 到 10
seed int 42 随机种子
temperature float 0.3 控制生成的温度参数
top_k int 20 生成文本时考虑的最高概率词的数量
top_p float 0.7 生成文本时的累积概率阈值
style string "" 风格
batch_size int 1 是否开启 batch 合成，小于等于 1 表示不使用 batch 合成（不推荐），范围是 1 到 20
spliter_threshold float 100 开启 batch 合成时，句子分割的阈值，范围是 10 到 1024
eos string "[uv_break]" 句子结束标志
enhance bool false 是否启用增强功能
denoise bool false 是否启用降噪功能

响应

返回一个音频文件，格式根据 response_format 参数确定。

示例

curl http://localhost:7870/v1/audio/speech \
  -H "Authorization: Bearer anything_your_wanna" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "chattts",
    "input": "Today is a wonderful day to build something people love! [lbreak]",
    "voice": "female2",
    "style": "chat"
  }' \
  --output speech.mp3

也可以使用 openai 库调用，具体可以看 openai 官方文档

Google API: /v1/text:synthesize

Endpoint

POST /v1/text:synthesize

Description

This endpoint synthesizes speech from text or SSML input using the Google Text-to-Speech API. It accepts various configuration parameters to customize the speech synthesis process. The API is designed to be compatible with Google's API while supporting additional parameters specific to this system.

Request Body

Field Type Description
input SynthesisInput Contains either text or ssml to be synthesized.
voice VoiceSelectionParams Specifies the voice parameters, including language code, speaker name, style, and other TTS options.
audioConfig AudioConfig Specifies the audio configuration, such as encoding format, speaking rate, pitch, and volume gain.
enhancerConfig EnhancerConfig (Optional) Additional configuration for audio enhancement.

SynthesisInput

Field Type Description
text str The text input to be synthesized.
ssml str The SSML input to be synthesized.

VoiceSelectionParams

Field Type Description
languageCode str The language code for the voice (default: "ZH-CN").
name str The name of the voice (default: "female2").
style str The style of the voice (default: "").
temperature float The temperature for the TTS model (default: 0.3).
topP float The topP value for the TTS model (default: 0.7).
topK int The topK value for the TTS model (default: 20).
seed int The seed value for inference (default: 42).
eos str The end-of-sentence marker (default: "[uv_break]").

AudioConfig

Field Type Description
audioEncoding AudioFormat The audio encoding format (default: AudioFormat.mp3).
speakingRate float The speaking rate (default: 1).
pitch float The pitch (default: 0).
volumeGainDb float The volume gain in dB (default: 0).
sampleRateHertz int The sample rate in Hertz (default: 24000).
batchSize int The batch size for inference (default: 4).
spliterThreshold int The splitter threshold for text splitting (default: 100).

EnhancerConfig

Field Type Description
... ... ...

Response

Field Type Description
audioContent str The base64-encoded audio content of the synthesis.

Example CURL Request

curl -X POST "http://localhost:7870/v1/text:synthesize" \
    -H "Content-Type: application/json" \
    -d '{
        "input": {
            "text": "Hello, this is a test."
        },
        "voice": {
            "languageCode": "ZH-CN",
            "name": "female2",
            "style": "",
            "temperature": 0.3,
            "topP": 0.7,
            "topK": 20,
            "seed": 42,
            "eos": "[uv_break]"
        },
        "audioConfig": {
            "audioEncoding": "mp3",
            "speakingRate": 1,
            "pitch": 0,
            "volumeGainDb": 0,
            "sampleRateHertz": 24000,
            "batchSize": 4,
            "spliterThreshold": 100
        },
        "enhancerConfig": null
    }'

XTTS v2 API

XTTS v2 系列 API 支持一些外部应用的调用，例如与 SillyTavern 的集成。

支持的 API 列表

方法 路径 描述
GET /v1/xtts_v2/speakers 获取可用的发音人列表
POST /v1/xtts_v2/tts_to_audio 将文本转化为音频文件
GET /v1/xtts_v2/tts_stream 获取 TTS 实时音频流
POST /v1/xtts_v2/set_tts_settings 设置 TTS 参数配置

具体的调用参数可以在启动服务后访问 /docs 端点查看。

如何连接到 SillyTavern？

通过 /v1/xtts_v2 系列 API，可以轻松将 ChatTTS-Forge 连接到 SillyTavern 中。以下是配置指南：

  1. 打开 SillyTavern 的插件扩展界面。
  2. 进入 TTS 插件的配置部分。
  3. 将 TTS Provider 切换为 XTTSv2。
  4. 勾选 Enabled 启用。
  5. 选择并配置 Voice 语音。
  6. 关键步骤：将 Provider Endpoint 设置为 http://localhost:7870/v1/xtts_v2。

sillytavern_tts

Forge Api v2

为了完全提供所有功能开发的 v2 接口，可以对所有可配置内容进行设置，同时还可以直接输入base64编码的音频作为参考音频创建临时 spk 用于推理

curl http://localhost:7870/v2/tts \
  -H "Authorization: Bearer anything_your_wanna" \
  -H "Content-Type: application/json" \
  -d '{
    "text": "Today is a wonderful day to build something people love!"
  }' \
  --output speech.mp3

usage

WIP

playground

启动 api 服务之后，在 /playground 下有一个非 gradio 的调试页面用于接口测试

playground

如果不需要此页面，启动服务的时候加上 --no_playground 即可