feat(tts): add Xiaomi MiMo-V2-TTS as speech provider by zoujiejun · Pull Request #55614 · openclaw/openclaw

zoujiejun · 2026-03-27T06:33:43Z

Summary

Add Xiaomi MiMo-V2-TTS as a new speech synthesis provider in the existing Xiaomi extension.

Closes #52376

Motivation

MiMo-V2-TTS provides high-quality Chinese TTS with unique capabilities:

Natural language style control — describe voice style in plain text
Dialect support — Northeastern, Sichuan, Cantonese, Taiwanese
Character voices — Sun Wukong, Lin Daiyu, etc.
Singing synthesis — same model, no mode switch
Free — currently the only free API on the MiMo platform

Implementation

Uses the existing Xiaomi extension. MiMo TTS uses a non-standard API:

Endpoint: POST /v1/chat/completions (not /v1/audio/speech)
Auth: api-key header (not Authorization: Bearer)
Audio: message.audio.data base64, 24kHz 16bit mono MP3

Configuration

{
  messages: {
    tts: {
      provider: "mimo",
      providers: {
        mimo: {
          apiKey: "sk-xxx",  // or MIMO_API_KEY env var
          voice: "mimo_default",
          style: "Happy"
        }
      }
    }
  }
}

Files Changed

File	Change
`extensions/xiaomi/speech-provider.ts`	New — 150 lines
`extensions/xiaomi/index.ts`	+2 lines
`extensions/xiaomi/openclaw.plugin.json`	+1 line

Testing

✅ pnpm build passes
✅ TTS tests 26/26 pass
✅ Real API: Chinese, English, style control, error handling, long text

AI-assisted (OpenClaw + GitHub Copilot). Fully tested.

greptile-apps · 2026-03-27T06:36:57Z

Greptile Summary

This PR adds Xiaomi MiMo-V2-TTS as a new speech synthesis provider within the existing Xiaomi extension. The implementation correctly follows the established SpeechProviderPlugin contract — auth fallback (config.apiKey || process.env.MIMO_API_KEY), fileExtension: \".wav\" (dot-prefixed as required by the TTS file-path template), and a static voice list are all consistent with the rest of the codebase.\n\nKey items:\n- The core speech-provider.ts implementation is clean and well-structured; no logic issues found.\n- extensions/xiaomi/index.ts correctly wires up the register hook.\n- bundled-plugin-metadata.generated.ts was not regenerated. openclaw.plugin.json now declares contracts.speechProviders: [\"mimo\"], but the auto-generated snapshot file still carries the old xiaomi entry without a contracts block. This causes pnpm check:bundled-plugin-metadata to fail and means BUNDLED_SPEECH_PLUGIN_IDS won't include \"xiaomi\" at runtime. Running node scripts/generate-bundled-plugin-metadata.mjs and committing the result will resolve both.

Confidence Score: 4/5

Safe to merge after regenerating and committing the bundled plugin metadata snapshot.

The speech provider implementation itself is correct and follows existing patterns. The single remaining blocker is the stale generated file — a one-command fix with no architectural change required.

src/plugins/bundled-plugin-metadata.generated.ts — needs to be regenerated to reflect the new contracts.speechProviders: ["mimo"] entry for the xiaomi plugin.

Important Files Changed

Filename	Overview
extensions/xiaomi/speech-provider.ts	New MiMo TTS speech provider implementing the non-standard chat-completions API. Structure, error handling, and auth fallback pattern all follow existing codebase conventions correctly.
extensions/xiaomi/index.ts	Minimal addition of the register hook that calls api.registerSpeechProvider(buildMimoSpeechProvider()). Correctly follows the same pattern used by other speech-provider plugins.
extensions/xiaomi/openclaw.plugin.json	Added contracts.speechProviders: ["mimo"] — correct intent, but the downstream generated file (bundled-plugin-metadata.generated.ts) was not regenerated, leaving the system's capability metadata stale and the CI check failing.

Prompt To Fix All With AI

This is a comment left during a code review.
Path: extensions/xiaomi/openclaw.plugin.json
Line: 3-5

Comment:
**Generated metadata file not regenerated**

`openclaw.plugin.json` was updated to add `contracts.speechProviders: ["mimo"]`, but the auto-generated file `src/plugins/bundled-plugin-metadata.generated.ts` was not regenerated. The xiaomi entry in that file still has no `contracts` key.

This has two concrete consequences:

1. `BUNDLED_SPEECH_PLUGIN_IDS` (computed from the generated file in `src/plugins/bundled-capability-metadata.ts:61`) will not include `"xiaomi"`, so the system has no record that the xiaomi plugin contributes a speech provider.
2. The `pnpm check:bundled-plugin-metadata` CI step (which runs `node scripts/generate-bundled-plugin-metadata.mjs --check`) will fail because the generated file no longer matches the actual plugin JSON on disk.

Run `node scripts/generate-bundled-plugin-metadata.mjs` from the repo root to regenerate the file and commit the result.

How can I resolve this? If you propose a fix, please make it concise.

_{Reviews (2): Last reviewed commit: "fix: correct plugin manifest contracts a..." | Re-trigger Greptile}

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 360f47083c

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 3e2982773f

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

Add MiMo-V2-TTS support as a new speech synthesis provider in the existing Xiaomi extension. MiMo TTS uses a non-standard endpoint (/v1/chat/completions with audio field) and authenticates via api-key header instead of Bearer token. Features: - Natural language style control via <style> tags - Three voices: mimo_default, default_zh, default_en - Support for dialects, role-playing, and singing synthesis - Currently free (no API cost) Configuration: messages.tts.provider: "mimo" messages.tts.providers.mimo.apiKey: "sk-xxx" Or set MIMO_API_KEY environment variable. Closes openclaw#52376

- voiceCompatible: false for WAV output (not in allowed voice-note formats) - regenerate bundled-plugin-metadata with speechProviders contract

MiMo TTS supports wav, mp3, pcm, pcm16 formats. Switch from wav to mp3: - MP3 is compatible with OpenClaw voice-note channels (Feishu, Telegram) - WAV caused voiceCompatible issues per src/media/audio.ts - MP3 is ~8x smaller (19KB vs 150KB for similar text)

MiMo TTS and text models share the same API platform. TTS provider now falls back to XIAOMI_API_KEY so users only need one key for both text inference and speech synthesis.

Co-authored-by: Copilot <[email protected]>

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 15437c22ec

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

chatgpt-codex-connector · 2026-03-30T11:45:45Z

+    aliases: ["xiaomi", "mimo"],
+    models: MIMO_TTS_MODELS,
+
+    isConfigured: ({ providerConfig }) =>
+      Boolean(


Normalize Xiaomi alias-backed TTS config lookup

This provider advertises xiaomi as an alias, but it does not normalize alias config keys to the canonical mimo id. In extensions/speech-core/src/tts.ts (resolveRawProviderConfig), TTS settings are resolved by canonical provider id, so messages.tts.providers.xiaomi.* is ignored and only messages.tts.providers.mimo.* is read; users who select provider: "xiaomi" and configure providers.xiaomi.apiKey will still hit not configured / MiMo API key missing unless an env var is present. Please add alias-aware resolveConfig handling (or remove the alias) so alias selection and config paths stay consistent.

Useful? React with 👍 / 👎.

steipete · 2026-04-25T07:50:58Z

Deep review against current main (5163a2f): feature is valid, but this PR needs work before land.

Positive evidence:

Current main has Xiaomi as a model provider only; no Xiaomi/MiMo speech provider is registered.
I live-tested with XIAOMI_API_KEY from .profile: POST https://api.xiaomimimo.com/v1/chat/completions with api-key auth, model: "mimo-v2-tts", and MP3 audio settings returned HTTP 200 with base64 audio.
The PR's single-assistant-message payload returned audio in the live probe. The issue's modalities: ["text", "audio"] + user/empty-assistant shape also returned audio.

Blockers:

CI is red on this head (checks-fast-extensions, multiple node test shards, Windows shards). Relevant enough that this needs a rebase/current green run before merge.
Generated bundled plugin metadata is stale after adding contracts.speechProviders: ["mimo"]; this breaks the capability metadata path and the generated metadata check.
No docs/changelog. Adding a public TTS provider needs docs/tools/tts.md, docs/providers/xiaomi.md, and a changelog entry.
No focused provider tests are in the diff. Please add tests for request shape, auth fallback (messages.tts.providers.mimo.apiKey -> MIMO_API_KEY/XIAOMI_API_KEY), error handling, base64 decode, outputFormat/fileExtension, and voice listing.
Consider accepting/documenting the explicit modalities: ["text", "audio"] request shape even though the current API accepted the shorter payload in my live probe. It is clearer and matches the confirmed working example in Feature Request: Add Xiaomi MiMo-V2-TTS as a native TTS provider #52376.

So: correct feature, API verified live, not solved on main yet.

steipete · 2026-04-25T08:48:37Z

Superseded by the implementation landed on main in ec8dbc4 (feat(tts): add xiaomi mimo speech provider).

I kept the feature from this PR, but rebuilt it with the missing maintainer pieces:

Provider id is xiaomi with mimo as an alias, matching the existing bundled Xiaomi plugin/provider naming.
The provider is Xiaomi-owned and declared in contracts.speechProviders.
Request shape follows Xiaomi's documented TTS contract: target text in an assistant message, optional style as a user message, api-key auth, and audio: { format, voice }.
Normal output is MP3/WAV; voice-note targets transcode to 48 kHz Opus with ffmpeg.
Added focused unit tests, live .profile tests, docs, and changelog.

Verification on the landed code:

pnpm check:changed
pnpm build
Xiaomi/MiniMax live TTS: 2 files / 5 tests passed with .profile keys.
Broader speech live sweep: OpenAI, ElevenLabs, Deepgram, Google, Gradium, and xAI live files: 6 files / 15 tests passed.

Closing this PR as superseded by the landed mainline fix. Thanks for pushing the Xiaomi TTS gap forward.

openclaw-barnacle Bot added extensions: xiaomi size: S labels Mar 27, 2026

zoujiejun marked this pull request as draft March 27, 2026 06:35

greptile-apps Bot reviewed Mar 27, 2026

View reviewed changes

Comment thread extensions/xiaomi/speech-provider.ts Outdated

chatgpt-codex-connector Bot reviewed Mar 27, 2026

View reviewed changes

Comment thread extensions/xiaomi/openclaw.plugin.json Outdated

Comment thread extensions/xiaomi/speech-provider.ts Outdated

zoujiejun marked this pull request as ready for review March 27, 2026 07:34

greptile-apps Bot reviewed Mar 27, 2026

View reviewed changes

Comment thread extensions/xiaomi/openclaw.plugin.json

chatgpt-codex-connector Bot reviewed Mar 27, 2026

View reviewed changes

Comment thread extensions/xiaomi/speech-provider.ts Outdated

Conan-Scott mentioned this pull request Mar 29, 2026

feat(fish-audio): add Fish Audio TTS speech provider #56891

Closed

24 tasks

zoujiejun and others added 6 commits March 30, 2026 19:22

fix: correct plugin manifest contracts and file extension format

c3115a6

fix: address bot review feedback

d98d51e

- voiceCompatible: false for WAV output (not in allowed voice-note formats) - regenerate bundled-plugin-metadata with speechProviders contract

fix: share XIAOMI_API_KEY between text model and TTS provider

9e3d20d

MiMo TTS and text models share the same API platform. TTS provider now falls back to XIAOMI_API_KEY so users only need one key for both text inference and speech synthesis.

extensions/xiaomi: fix MiMo TTS formatting

15437c2

Co-authored-by: Copilot <[email protected]>

zoujiejun force-pushed the feat/mimo-tts-provider branch from bd74745 to 15437c2 Compare March 30, 2026 11:39

chatgpt-codex-connector Bot reviewed Mar 30, 2026

View reviewed changes

steipete mentioned this pull request Apr 25, 2026

Feature Request: Add Xiaomi MiMo-V2-TTS as a native TTS provider #52376

Closed

steipete closed this Apr 25, 2026

This was referenced Apr 30, 2026

feat(tts): support MiMo v2.5 TTS VoiceDesign #72927

Closed

fix(tts/xiaomi): Xiaomi Token Plan endpoint uses Bearer auth, not api-key header #77692

Open

clawsweeper Bot mentioned this pull request May 8, 2026

fix(tts/xiaomi): support Token Plan TTS endpoint #79185

Open

clawsweeper Bot mentioned this pull request May 15, 2026

fix(tts/xiaomi): support per-call voice and model overrides #77792

Closed

clawsweeper Bot mentioned this pull request May 28, 2026

feat(tts): xiaomi voicedesign/voiceclone model support #87816

Open

clawsweeper Bot mentioned this pull request Jun 26, 2026

fix(speech): bound TTS response reads #96874

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Uh oh!

feat(tts): add Xiaomi MiMo-V2-TTS as speech provider#55614

feat(tts): add Xiaomi MiMo-V2-TTS as speech provider#55614
zoujiejun wants to merge 6 commits into
openclaw:mainfrom
zoujiejun:feat/mimo-tts-provider

zoujiejun commented Mar 27, 2026 •

edited

Loading

Uh oh!

greptile-apps Bot commented Mar 27, 2026 •

edited

Loading

Important Files Changed

Uh oh!

Uh oh!

chatgpt-codex-connector Bot left a comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

chatgpt-codex-connector Bot left a comment

Uh oh!

Uh oh!

chatgpt-codex-connector Bot left a comment

Uh oh!

chatgpt-codex-connector Bot Mar 30, 2026

Uh oh!

steipete commented Apr 25, 2026

Uh oh!

steipete commented Apr 25, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Uh oh!

Conversation

zoujiejun commented Mar 27, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Motivation

Implementation

Configuration

Files Changed

Testing

Uh oh!

greptile-apps Bot commented Mar 27, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Greptile Summary

Confidence Score: 4/5

Important Files Changed

Uh oh!

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

Uh oh!

Uh oh!

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

chatgpt-codex-connector Bot Mar 30, 2026

Choose a reason for hiding this comment

Uh oh!

steipete commented Apr 25, 2026

Uh oh!

steipete commented Apr 25, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

zoujiejun commented Mar 27, 2026 •

edited

Loading

greptile-apps Bot commented Mar 27, 2026 •

edited

Loading