audiosr.md

AudioSR

AudioSR performs audio super-resolution from an input waveform.

Quick Start

audiocpp_cli \
  --task s2s \
  --family audiosr \
  --model models/AudioSR-GGUF/audiosr-basic-f32.gguf \
  --backend cuda \
  --audio input.wav \
  --request-option num_inference_steps=50 \
  --request-option guidance_scale=3.5 \
  --request-option ddim_eta=1.0 \
  --request-option seed=42 \
  --out output.wav

Model

Field	Value
Family	audiosr
Task	s2s
Mode	offline
Default package	models/AudioSR-GGUF/audiosr-basic-f32.gguf
Input	Source WAV through --audio
Output	Super-resolved waveform

ddim_eta defaults to 1.0, matching the official AudioSR inference default. Use --request-option ddim_eta=0 only for deterministic sampler debugging.

Long audio is processed with bounded overlapping chunks once the input exceeds audio_chunk_duration_sec. The defaults match the official Python long-audio path:

--request-option audio_chunk_duration_sec=15 \
--request-option audio_chunk_overlap_sec=2

Options

Option	Values	Default	Meaning
--audio	WAV path	required	Source waveform to super-resolve.
--request-option num_inference_steps=<n>	integer > 0	50	DDIM sampler steps.
--request-option guidance_scale=<f>	float	3.5	Classifier-free guidance scale.
--request-option ddim_eta=<f>	float >= 0	1.0	DDIM eta.
--request-option audio_chunk_duration_sec=<seconds>	seconds	15	Long-audio chunk size.
--request-option audio_chunk_overlap_sec=<seconds>	seconds	2	Overlap between long-audio chunks.
--request-option seed=<n>	integer, -1 for random	42	Sampler seed.
--session-option audiosr.weight_type=<type>	native, f32, f16, bf16, q8_0	native	Weight storage override for experiments.