Piper TTS PowerShell Wrapper

A production-ready PowerShell toolchain for Speech Synthesis (TTS). Orchestrates Piper for inference and FFmpeg for encoding, optimized for high-density audio asset generation.

Technical Features

• Dual-Stage Pipeline: Implements a WAV-to-Opus workflow, reducing storage requirements by up to 15x while maintaining vocal clarity.
• Batch Processing: Automated high-throughput synthesis from plaintext datasets.
• Manifest Generation: Exports a map.json linking source text to generated artifacts for database or Anki integration.
• Encoding Stability: Enforces UTF-8 console and input handling for multilingual support (German, Polish, etc.).

Setup

1. Requirements

• Piper: Download Windows binary from Piper GitHub Releases.
• FFmpeg: Required for Opus encoding. Download from ffmpeg.org.
• System PATH: Ensure both piper and ffmpeg are added to your Environment Variables.

2. Voice Models

Download .onnx and .json files from the Piper Voices Repository and place them in your $voiceDir.

Usage

Individual Synthesis (say.ps1)

.\say.ps1 -text "Ich möchte ein Studienkolleg besuchen" -lang "de" -speed 1.3

Bulk Processing (mass_say.ps1)

Expects a UTF-8 encoded .txt file with one phrase per line:

.\mass_say.ps1 -inputFile "vocab.txt" -lang "de" -speed 1.2

Parameters

• -text / -inputFile: The content to convert (Mandatory).
• -lang: Language code (en, de, dem (emotional), pl). Defaults to de.
• -speed: The length_scale factor. Higher is slower (e.g., 1.5). Defaults to 1.2.

Output

Files are saved to your $outputDir in .opus format. Mass processing generates a map.json:

[
  {
    "id": 1,
    "text": "Guten Tag",
    "audio": "[sound:de_audio_1.opus]"
  }
]

Troubleshooting

• Encoding: Ensure input files are UTF-8 encoded without BOM to prevent character corruption.
• FFmpeg: If encoding fails, verify FFmpeg is accessible via terminal and you have write permissions for the destination directory.