A simple, containerized web API for real-time text-to-speech (TTS) synthesis using Piper, powered by FastAPI.

• 📥 Accepts text input and returns a synthesized speech .wav file
• 🧹 Automatically deletes audio files older than 1 hour
• 🎙️ Supports multiple languages and voice models (.onnx)
• 🪄 Apply custom audio effects via JSON-defined effect chains
• 🎛️ Effects include: pitch_shift, normalize, flanger, random_semitone_sawtooth_wave — with full parameter control
• 💾 Optionally convert output to portable WAV format:
  • Mono (1 channel)
  • 16-bit sample width
  • 48,000 Hz sample rate
• 🔐 Optional API key protection with HTTP Basic Auth (via API_KEY env variable)
• 📡 /api/healthcheck endpoint is always public for container monitoring
• 🐳 Built to run easily in Docker with configurable ports and bind-mounted output
• ⚡ Fast, lightweight, production-ready thanks to FastAPI + Piper TTS

Synthesize speech from text with optional effects.

{
  "text": "Hello, world!",
  "local": "fr_FR",                    // Optional - default: "fr_FR"
  "voice": "siwis-medium",             // Optional - default: "siwis-medium"
  "silence": 1,                        // Optional - sentence silence (seconds)
  "speed": 1.0,                        // Optional - speech speed factor
  "noise_w": 0.8,                      // Optional - noise weight
  "effects": [                         // Optional - list of effect steps to apply
    {
      "name": "pitch_shift",
      "params": { "pitch_change": 10 }
    },
    {
      "name": "random_semitone_sawtooth_wave",
      "params": { "min_freq": 170, "max_semitones": 6, "pitch_duration": 0.4, "wet": 0.3 }
    },
    {
      "name": "normalize",
      "params": {}
    }
  ],
  "lite_file": true                    // Optional - convert to 16-bit mono WAV
}

{
  "filename": "abcd1234.wav",
  "url": "/api/v1/synthesize/abcd1234.wav"
}

Download a previously synthesized .wav file.

Simple healthcheck for the server.

You can protect your API with Basic HTTP Authentication by setting the API_KEY environment variable.

• If API_KEY is not set (default): all /api/* endpoints are publicly accessible.

• If API_KEY is set, then:

  • All /api/* routes require Basic Auth:

    • Username: piper
    • Password: your API_KEY value

  • The /api/healthcheck endpoint remains public for monitoring purposes.

Set the environment variable when running the server:

export API_KEY=mysecurekey
uvicorn piper_tts_server:app --host 0.0.0.0 --port 8000

Then access the API like this:

curl -u piper:mysecurekey -X POST http://localhost:8000/api/v1/synthesize \
  -H "Content-Type: application/json" \
  -d '{ "text": "Bonjour !", "lite_file": true }'

If authentication fails, you'll receive:

{
  "detail": "Unauthorized"
}

services:
  piper-tts:
    image: arkdevuk/webpiper:latest
    container_name: piper-tts
    ports:
      - "8080:8080" # exposed port : container port, you can change the exposed port if needed
    volumes:
      - ./output:/output         # Mount output directory to access generated audio
    restart: unless-stopped
    environment:
      - API_KEY: mysecurekey # Optional - set API key for authentication

The API will be available at: http://localhost:8080

You can chain multiple effects in the effects parameter. Each effect has a name and optional params field.

Example:

{
  "effects": [
    { "name": "pitch_shift", "params": { "pitch_change": 10 } },
    { "name": "normalize", "params": {} }
  ]
}

You can define a list of effects using the effects parameter in the request body. Each effect should have a name and an optional params dictionary.

Applies a metallic-sounding flanger modulation.

Parameters:

• rate (float): LFO rate in Hz. Controls how fast the pitch shifts. Default: 0.15
• min_delay (float): Minimum delay in seconds. Sets the starting point for the flanging. Default: 0.0025
• max_delay (float): Maximum delay in seconds. Sets the depth of the flanging. Default: 0.0035
• feedback (float): Amount of feedback (0 to 1). Higher values produce more intense echoes. Default: 0.9
• t_offset (float): Static time offset for the modulation phase. Default: 0
• dry (float): Dry (original) signal level. Range: 0 to 1. Default: 0.5
• wet (float): Wet (effected) signal level. Range: 0 to 1. Default: 0.5

Removes DC offset and scales the audio to maximum amplitude.

Parameters:

• (none)

Note: Automatically centers the waveform and scales it to full 16-bit range.

Changes the pitch of the audio without altering its duration.

Parameters:

• pitch_change (int): Pitch shift amount in semitone percent. Range: -100 (one octave down) to +100 (one octave up). Default: 0

Applies amplitude modulation using a sawtooth wave with random semitone-based frequency changes.

Parameters:

• min_freq (float): Minimum frequency of the base sawtooth wave in Hz. Example: 170.0
• max_semitones (int): Max number of semitones to shift up from min_freq. Higher = more variation.
• pitch_duration (float): Duration (in seconds) of each pitch modulation segment.
• wet (float): Strength of the effect (0 = no effect, 1 = full modulated signal). Default: 0.5

Changes the playback speed of the audio by resampling.

Parameters:

• speed (float): Speed adjustment factor.
  • 0.0 = no change
  • Positive values speed up the audio (0.25 = 25% faster)
  • Negative values slow it down (-0.25 = 25% slower)
  • Range: around -0.9 to +1.0 is typical for experimental use

⚠️ This effect does alter the duration of the audio. It’s a simple and accurate time-domain resampling.

Piper voice files (.onnx) are stored in /voices.

You can find voice models here: 👉 https://github.com/rhasspy/piper/blob/master/VOICES.md

Download and place them inside the voices/ folder. They should be named like: fr_FR-siwis-medium.onnx.

• Python 3.10
• FastAPI
• Uvicorn
• Piper TTS Engine

Install with:

pip install -r requirements.txt

MIT License

• Piper - Lightweight, fast TTS engine
• FastAPI - High performance web framework
• Voicebox - Inspiration to build the effects feature