Piper TTS (en_US lessac medium) — GGUF

Native C++ GGUF conversion of the Piper VITS voice en_US-lessac-medium for use with CrispASR.

Files

Piper models are small enough that quantization provides no meaningful savings. F16 is the only format.

Usage with CrispASR

./build/bin/crispasr --backend piper \
    -m piper-en_US-lessac-medium-f16.gguf \
    --tts "Hello, how are you today?" \
    --tts-output hello.wav

Phonemization uses espeak-ng (must be installed: apt install espeak-ng).

Architecture

• VITS (Conditional Variational Autoencoder with Adversarial Learning)
• Text encoder: 6-layer relative-position transformer (192-d, 2 heads)
• Duration predictor: Stochastic Duration Predictor with rational-quadratic spline flows
• Flow: 4 affine coupling blocks with WaveNet conditioning
• Decoder: HiFi-GAN (3 upsample stages, 9 MRF resblocks)
• Output: 22.05 kHz mono PCM
• License: MIT

Conversion

python models/convert-piper-to-gguf.py \
    --onnx en_US-lessac-medium.onnx \
    --output piper-en_US-lessac-medium-f16.gguf

Provenance and EU AI Act Art. 53 note

• Upstream model: rhasspy/piper-voices — published by rhasspy.
• Upstream licence: mit. This repository redistributes under the same terms; it grants no rights the upstream licence does not.
• What was done here: format conversion and/or quantisation only (GGUF). No training, no fine-tuning, no merging, no distillation, no change to architecture, vocabulary or capability. Only the numeric representation of the upstream weights differs.
• Training data: documented — where it is documented at all — by the upstream provider; see the upstream model card. No training data was used, added or selected by this repository. No training-content summary was found on the upstream model card at the time of writing; that documentation gap is upstream's and is not filled here.
• Provider status: under Regulation (EU) 2024/1689 the upstream authors remain the provider of this model. Converting the serialisation format does not make this repository the provider of a new general-purpose AI model, and no such claim is made. Questions about training content, copyright policy or model capability belong upstream.