docs: fix ASR content-type warning + document WS HTTP/1.1 handshake

[M2Night]

What

Two documentation fixes found while helping Vercel build a Fish Audio provider for the AI SDK.

Changes

• Speech to Text warning (speech-to-text.mdx): application/form-data is not a real MIME type → multipart/form-data; also note application/json is not usable (the audio field cannot be carried as JSON/base64). Verified: JSON+base64 → HTTP 400; multipart/form-data → 200.
• WebSocket TTS (asyncapi.yml): document that the handshake must use HTTP/1.1 — HTTP/2 upgrade requests are rejected with 400 "invalid protocol". Verified live (curl --http1.1 → 101; default HTTP/2 → 400 invalid protocol).

Note

The /v1/asr OpenAPI content-type list itself is corrected in the companion platform-api PR (that repo generates openapi.json, synced here via scripts/update-openapi.mjs).

🤖 Generated with Claude Code

---

^{Need help on this PR? Tag @codesmith-bot with what you need. Autofix is disabled.}

Summary by CodeRabbit

• Documentation
  • Added WebSocket handshake requirements clarifying that HTTP/1.1 protocol must be used and HTTP/2 clients will be rejected with a 400 error.
  • Corrected file upload MIME type documentation to multipart/form-data with clarification that application/json is not accepted for audio data submission.

[mintlify]

Preview deployment for your docs. Learn more about Mintlify Previews.

Project	Status	Preview	Updated (UTC)
hanabiaiinc	🟢 Ready	View Preview	Aug 4, 2026, 3:50 PM

💡 Tip: Enable Workflows to automatically generate PRs for you.

[coderabbitai]

📝 Walkthrough

Walkthrough

The API documentation now defines the HTTP/1.1 requirement for ttsLive WebSocket handshakes and corrects Speech to Text content-type guidance.

Changes

API documentation updates

Layer / File(s)	Summary
WebSocket handshake requirements api-reference/asyncapi.yml	Documents the HTTP/1.1 requirement and the 400 invalid protocol response for HTTP/2 clients.
Speech to Text request format api-reference/endpoint/openapi-v1/speech-to-text.mdx	Corrects the MIME type to multipart/form-data and states that application/json with JSON/base64 audio is not accepted.

Estimated code review effort: 1 (Trivial) | ~3 minutes

Possibly related PRs

• fishaudio/docs#24: Updates the same api-reference/asyncapi.yml TTS WebSocket specification.

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly and concisely summarizes both documentation changes: the ASR content type correction and the WebSocket HTTP/1.1 handshake requirement.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch vercel-asr-tts-doc-fixes

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[coderabbitai]

Actionable comments posted: 1

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@api-reference/endpoint/openapi-v1/speech-to-text.mdx`:
- Line 10: Update the endpoint description to explicitly state that it rejects
JSON request bodies containing base64-encoded audio, while preserving the listed
accepted content types and avoiding the ambiguous “JSON/base64” phrasing.

🪄 Autofix

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: 133e26a3-9998-4324-9f24-e2f8a865d17a

📥 Commits

Reviewing files that changed from the base of the PR and between fc0fcb8 and a454133.

📒 Files selected for processing (2)

• api-reference/asyncapi.yml
• api-reference/endpoint/openapi-v1/speech-to-text.mdx

[coderabbitai]

🎯 Functional Correctness | 🟡 Minor | ⚡ Quick win

Clarify the JSON restriction.

JSON/base64 combines a media type and an encoding. State that the endpoint rejects JSON bodies with base64-encoded audio.

Proposed wording

-This BETA endpoint only accepts `multipart/form-data` and `application/msgpack`. It does not accept `application/json` (the audio field cannot be carried as JSON/base64).
+This BETA endpoint only accepts `multipart/form-data` and `application/msgpack`. It does not accept `application/json` with base64-encoded audio.

📝 Committable suggestion

‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.

Suggested change

	This BETA endpoint only accepts `multipart/form-data` and `application/msgpack`. It does not accept `application/json` (the audio field cannot be carried as JSON/base64).
	This BETA endpoint only accepts `multipart/form-data` and `application/msgpack`. It does not accept `application/json` with base64-encoded audio.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@api-reference/endpoint/openapi-v1/speech-to-text.mdx` at line 10, Update the
endpoint description to explicitly state that it rejects JSON request bodies
containing base64-encoded audio, while preserving the listed accepted content
types and avoiding the ambiguous “JSON/base64” phrasing.

Source: Coding guidelines