Piper Fallback TTS

Kokoro is the production audio path for this project. Piper is kept as a durable fallback and comparison engine for machines where Kokoro is unavailable or when maintainers want to compare voice quality.

When to Use Piper

Use Piper only when:

• Kokoro dependencies cannot be installed on the target machine.
• You need a local fallback to compare pronunciation, pacing, or voice character.
• You are diagnosing whether an issue is script-related or engine-related.

Do not use Piper for the final published audio unless the project intentionally switches production engines.

Setup

Run these commands from the repository root:

python -m pip install piper-tts
python -m podcasts.tts.download_samples

The downloaded Piper models live under podcasts/tts/models/, which is intentionally ignored by git.

Generate Audio with Piper

Use the npm wrapper:

npm run build:podcast-audio:piper

Or call the unified audio command directly:

python -m podcasts.tts.generate_audio --engine piper --start 0 --end 10 --audio-format mp3

The older Piper module remains available for diagnostics:

python -m podcasts.tts.generate_all --start 0 --end 10

The legacy Piper batch path follows the same canonical listening order as Kokoro, so Challenge Coach episodes stay near the chapters they support.

Configuration

Piper runtime settings live in podcasts/tts/voice-config.ini and can also be overridden with environment variables documented in generate_episode.py.

The default Piper voices are:

Host	Piper model	Purpose
Alex	en_US-ryan-high.onnx	Lead host fallback voice
Jamie	en_US-lessac-high.onnx	Co-host fallback voice

Validation

After a Piper test run, use the same checks as Kokoro:

npm run podcast:inventory:check
npm run podcast:metadata:check
npm run build:podcast-site
npm run validate:podcast-feed

For production, return to Kokoro unless maintainers explicitly decide otherwise.