Text-to-speech · TTS

Piper

Open-weight, efficient TTS — multilingual and CPU-friendly.

Piper is the cheapest voice in our catalog. Originally built for Home Assistant and edge devices, it is small, efficient, and supports a broad multilingual voice catalog. Quality is below Chatterbox and Kokoro for English, but acceptable, and the language coverage is wider.

When to pick it

  • You need voice in a language Chatterbox or Kokoro does not cover
  • Volume is high enough that cost-per-character dominates
  • The output is going to be heard once and discarded (notifications, one-off briefings)

When to pick something else

Pricing

EUR 2.00 per million characters. The price floor of the catalog — roughly half of AWS Polly Standard ($4/M) and well under every neural-voice tier.

Limits

Same operational limits as the other TTS models. Voice catalog browseable through the API; specific voice availability per language varies.

Best for

  • High-volume notification pipelines where cost per minute matters
  • Multilingual voice output across many smaller languages
  • Background voice generation in non-real-time pipelines
  • Acceptable-quality voice at the floor of the pricing curve

Upstream source: github.com/rhasspy/piper