---
title: FLowHigh
emoji: 📶
colorFrom: green
colorTo: indigo
sdk: gradio
sdk_version: 6.24.0
python_version: '3.10'
preload_from_hub:
  - ResembleAI/FlowHigh
app_file: app.py
pinned: false
license: mit
short_description: Flow-matching audio bandwidth extension
tags:
  - category:enhancement
  - subcategory:bandwidth-extension
  - domain:speech
  - input:audio
  - output:audio
  - sample-rate:48000
models:
  - ResembleAI/FlowHigh
---

FLowHigh audio bandwidth extension for HARP.

Upload an audio clip of up to 30 seconds. Stereo inputs are mixed to mono.
The model runs at 48 kHz; 24 and 96 kHz outputs are resampled. Selecting
96 kHz does not extend bandwidth beyond the model's 48 kHz result.
Downloads are available for one hour.

Runs on ZeroGPU or a dedicated NVIDIA GPU. Weights download automatically
from `ResembleAI/FlowHigh` at startup.

HARP endpoints: `/controls`, `/process`, `/cancel`.
Processing takes an audio file and an output-rate string: `"24000"`,
`"48000"`, or `"96000"`.

See [SOURCES.md](SOURCES.md) for source repositories and revisions.