Map Resemble Enhance compatibility feasibility

[Tinnci]

Goal

Map whether Resemble Enhance should be added as a speech-focused external backend, self-contained compatibility target, or documentation-only recommendation.

Source facts to verify

• Repository: https://github.com/resemble-ai/resemble-enhance
• License: MIT
• Package: resemble-enhance
• Domain: speech denoising and enhancement/bandwidth extension
• Training target: high-quality 44.1 kHz speech
• Enhancer path uses latent conditional flow matching and large checkpoint files.
• Upstream download metadata references Hugging Face repository ResembleAI/resemble-enhance.

Acceptance criteria

• Identify the minimal files needed for enhancer-only inference.
• Record checkpoint sizes, hashes, license, and provider metadata.
• Decide whether deepspeed/training dependencies can be avoided for inference.
• Verify local array/file I/O behavior on a short speech fixture if dependencies are available.
• Decide whether this belongs as an external package backend, self-contained compat backend, or deferred candidate.

Non-goals

• Do not implement the backend in this issue.
• Do not add Resemble Enhance dependencies to the default package.
• Do not blur speech-only enhancement with general-audio SR catalog entries.