Skip to content
← Models
Open weightsResemble AI/10 trustVerified

Resemble Enhance

Two-stage open-source speech restorer: a denoiser that separates speech from noise, then an enhancer that repairs distortion and extends bandwidth — pip install and go.

Visit official page →

About

Resemble Enhance is the easiest serious speech-restoration tool in this batch to actually run. It chains two models: a denoiser that separates speech from a noisy recording, and an enhancer that goes further — repairing distortion and extending the audio bandwidth, trained on high-quality 44.1 kHz speech so the output isn't telephone-grade. Installation is one pip command, there's a CLI for batch folders, a --denoise_only flag if you want just the first stage, and a Gradio web demo on Hugging Face Spaces. Resemble AI released it under MIT in late 2023, and at 2.4k stars it's become a default 'clean up this podcast/voiceover' utility in open-source workflows. The honest limits: there's no published benchmark paper, so quality claims rest on the company's examples and community experience rather than numbers; it's GPU-hungry for long files; and like all generative-ish enhancers it can smooth over or invent detail on inputs far from its training distribution. But for a free, one-command restorer with permissive licensing, it's the pragmatic pick.

Specs

Context window

Not applicable — audio file in, restored audio file out (offline)

Pricing

Free / open-source (MIT); no hosted paid tier for this tool

Free quota

Fully open — self-host, no usage limits

API access

No hosted API — pip install resemble-enhance; checkpoints download automatically; Gradio web demo on Hugging Face Spaces

Weights

Open weights

Released

Not disclosed

Benchmarks

  • No published benchmarks or paper — quality judged from company demos and community use
  • Provider-claimed: trained on high-quality 44.1 kHz speech data

Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.

Use cases

speech-enhancementaudio-restorationpodcast-productionvoiceover-cleanupbandwidth-extension

Trust score — 8/10

/10 trustVerified

Scored 8/10 (TIER verified, metrics-v1). Evidence: public MIT-licensed repo from Resemble AI (2.4k stars) with pip-installable checkpoints, an HF Spaces demo, and sustained community use since Nov 2023. Strengths: genuinely easy to deploy, permissive license, two-stage design covers denoise + restoration + bandwidth extension. Docked for: (1) no published benchmarks or paper — there is no numbered evidence of quality, only demos and word of mouth; (2) heavy on GPU for long audio. License: MIT (Copyright 2023 Resemble AI).

How the trust score is built

Full methodology →
Adoption · stars/downloads, log-scaled vs peers
7.0/10
Maintenance · recency, releases, issue load
7.0/10
Openness · license + docs
10.0/10
Manual review · sanity read + red-flag check
8.0/10
  • · resemble-ai/resemble-enhance repo created 2023-11-15, ~2.4k stars
  • · Two-stage: denoiser + enhancer (distortion repair + bandwidth extension), trained on high-quality 44.1 kHz speech
  • · No published benchmarks or paper; pip install + CLI + HF Spaces Gradio demo

Sources