dac-44khz-GGUF

GGUF conversion of descript/dac_44khz — the Descript Audio Codec (DAC) decoder at 44.1 kHz.

Used as the vocoder backend for Zonos-v0.1-transformer-GGUF and other DAC-based TTS models running in CrispASR.

Files

File	Size	Description
dac-44khz-f16.gguf	104 MB	F16 weights, full precision

Architecture

The DAC decoder converts 9 residual VQ codebooks (1024 entries each, ~86 codes/s) to 44.1 kHz mono PCM:

• RVQ quantizer: 9 codebooks × 1024 entries × 8-dim → projected to 1024-dim
• 4-stage convolutional decoder with Snake1d activations and upsampling ratios [8, 8, 4, 2]
• Total upsampling factor: 512 → 86 Hz codes × 512 = 44,032 Hz ≈ 44.1 kHz

Usage

Place alongside the Zonos transformer GGUF or pass via --codec-model:

# Auto-discovered if named dac-44khz-f16.gguf in the same directory as the AR model
crispasr tts --backend zonos-tts \
  --model /path/to/zonos-v0.1-transformer-q4_k.gguf \
  --codec-model /path/to/dac-44khz-f16.gguf \
  "Hello world."

Conversion

Converted from descript/dac_44khz using models/convert-dac-to-gguf.py in CrispASR. Source model is MIT licensed (descriptinc/descript-audio-codec).

Provenance and EU AI Act Art. 53 note

• Upstream model: descript/dac_44khz — published by descript.
• 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.
• 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.