docs: surface merged speech-to-speech SenseVoice integration

[LauraGPT]

Production Publication (2026-10-01, Asia/Shanghai)

The tested source in this PR is now published at www.funasr.com as release 20260930T211757Z (verified 2026-09-30T21:57:20Z). Chinese guide / English guide.

Production validation covered all 273 file hashes and preserved all 206 existing HTML routes. Thirteen public routes returned HTTP 200 with exact body hashes and security headers. A subsequent public-domain Chromium run passed four Chinese/English desktop/mobile card-to-guide cases. Fresh backup and previous release were preserved; the existing Nginx master was gracefully reloaded with unchanged identity/configuration. No model service restart, model inference or Python-package release.

The operational record is tracked in #3745. The no-deployment statements below describe the earlier source-merge stage, before this subsequent production publication.

Summary

• Surface the upstream-merged Hugging Face speech-to-speech SenseVoice integration in both README files, the bilingual community guide, and both website ecosystem pages.
• Pin the source installation to upstream merge 9e2ed1099190a4e4bc8a972b4a3488949ff1b9f6. Explicitly distinguish this source integration from the released v1.0.0, which does not contain it.
• Document the actual VAD-segment/non-streaming model boundary, separate LLM/TTS setup, initial model downloads, and terminal transcript display versus application logging.
• Replace the obsolete pending-review growth entry with the verified merged status. Refresh only the two affected generated legacy checksums.

Validation

• New regressions: 3 failed before the content change, 3 passed afterward.
• Full product-site suite: 521 passed.
• Selected root documentation tests: 100 passed, 1 skipped. The skipped module is tests/test_sphinx_links.py, because Sphinx is absent in the validation environment.
• Repeated the full site suite and selected documentation checks after fast-forwarding to actual main be7557f27ccc42662b2b55236a0f7d2d67661813; documentation and website source hashes remained identical.
• Two independent builds each generated 117 product pages; both validators checked 206 total pages; recursive output comparison was identical.
• Actual Playwright Chromium checks: four cases, EN/ZH at desktop 1365px and mobile 390px. Verified unique ecosystem entry, no horizontal overflow, localized guide link/anchor, pinned recipe, and no JavaScript errors. External requests were blocked; screenshots inspected.
• git diff --check passed. Independent read-only review found no P1/P2 issues on the tested source hashes.

This is documentation and website-source work, not a runtime change, model/GPU acceptance, full voice-conversation test, new upstream CLI validation, PyPI release, or production website deployment. Historical upstream test evidence is not claimed as a fresh run here.

Upstream

huggingface/speech-to-speech#319 was approved and merged by the upstream maintainer on 2026-09-29 UTC.

Integration Status (2026-10-01, Asia/Shanghai)

Merged as 460f00675499924fec0a14bdbeb7125ed505dbb2. The merge tree, parents, GitHub signature and all ten changed-file SHA-256 values match the tested branch.

Exact-head hosted checks:

• Test llama.cpp downloader: success.
• Validate repository roadmap: success.
• Validate MOSS adapter: success.
• Product site: success.

The Product site workflow includes successful legacy-document link, build/validation and browser jobs. This hosted link check is separate from the local Sphinx skip documented above.

Exact-merge checks at this observation:

• Validate MOSS adapter: completed / success.
• Product site: completed / success.
• Test llama.cpp downloader: completed / success.
• Validate repository roadmap: completed / success.
• Update API Documentation: completed / success.

Used the ordinary SHA-guarded merge API under the repository's existing administrator exemption. No protection/rules changes, invented human approval, workflow approval, force push, package release or production deployment.