Greptile Summary

This PR adds Volcengine (ByteDance/豆包) as a TTS speech provider, integrating it into the existing volcengine plugin alongside the LLM provider. The implementation is well-structured, follows existing provider conventions, and includes type definitions, Zod schema validation, and unit tests.

Key changes:

• New extensions/volcengine/tts.ts — raw HTTPS client for the ByteDance Speech API, returning base64-decoded audio buffers
• New extensions/volcengine/speech-provider.ts — SpeechProviderPlugin implementation with 6 built-in Chinese voices, OGG Opus/MP3 encoding, and credential resolution from config or env vars
• Core type updates in src/config/types.tts.ts, src/config/zod-schema.core.ts, and src/tts/tts.ts

Issues found:

• P1 — Missing res.on("error") handler in tts.ts: Response-stream errors are not caught. If the IncomingMessage emits "error" (e.g. a mid-stream TCP reset), the Node.js process receives an uncaught exception and the returned Promise hangs forever since neither resolve nor reject is called.
• P2 — req referenced in timer callback before const declaration: Technically safe at runtime but unconventional and may trigger linter warnings.
• P2 — Unused imports in tts.test.ts: vi, beforeEach, and afterEach are imported but never used.

Confidence Score: 4/5

Safe to merge after fixing the missing response error handler, which can cause unhandled exceptions on network failures.

The overall implementation is clean, well-typed, and consistent with the existing codebase. The only actionable bug is the missing res.on('error') listener in the HTTP client, which is a reliability gap on network fault paths. The remaining comments are style-level. One targeted fix resolves the primary concern.

extensions/volcengine/tts.ts — needs the response stream error handler before merge.

This is a comment left during a code review.
Path: extensions/volcengine/tts.ts
Line: 73-91

Comment:
**Missing `res.on("error")` handler leaks promise**

The response stream (`res`) has no `"error"` listener. In Node.js, if an `IncomingMessage` stream emits `"error"` (e.g. due to a mid-stream connection reset), the event has no handler, which causes an uncaught exception in the process. The outer `req.on("error")` only covers request-side errors, not response-side stream errors. The promise would also hang indefinitely because neither `resolve` nor `reject` would be called.

Add an `"error"` listener on `res`:

```suggestion
      (res) => {
        const chunks: Buffer[] = [];
        res.on("data", (chunk: Buffer) => chunks.push(chunk));
        res.on("error", (e) => {
          clearTimeout(timer);
          reject(new Error(`Volcengine TTS response error: ${e.message}`));
        });
        res.on("end", () => {
          clearTimeout(timer);
          try {
            const body = JSON.parse(Buffer.concat(chunks).toString());
            if (body.code === 3000 && body.data) {
              resolve(Buffer.from(body.data, "base64"));
            } else {
              reject(
                new Error(`Volcengine TTS error ${body.code}: ${body.message ?? "unknown"}`),
              );
            }
          } catch (e) {
            reject(new Error(`Volcengine TTS: failed to parse response: ${e}`));
          }
        });
      },
```

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

---

This is a comment left during a code review.
Path: extensions/volcengine/tts.ts
Line: 57-61

Comment:
**`req` captured in timer closure before declaration**

`req` is referenced inside the `setTimeout` callback on line 58, but it's only declared with `const` on line 62. This works at runtime because `setTimeout` is asynchronous and `req` will always be set before the timer fires. However, it creates a temporal dead zone reference that some linters (e.g. `no-use-before-define`) may flag, and it's easy to misread.

Consider swapping the ordering — assign `req` first, then create the timer:

```typescript
const req = https.request({ ... }, (res) => { ... });
const timer = setTimeout(() => {
  req.destroy();
  reject(new Error(`Volcengine TTS timeout after ${timeoutMs}ms`));
}, timeoutMs);
req.write(payload);
req.end();
```

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

---

This is a comment left during a code review.
Path: extensions/volcengine/tts.test.ts
Line: 1

Comment:
**Unused imports: `vi`, `beforeEach`, `afterEach`**

`vi`, `beforeEach`, and `afterEach` are imported from `vitest` but never used in the test file. These can be removed to avoid linting warnings.

```suggestion
import { describe, it, expect } from "vitest";
```

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

_{Reviews (1): Last reviewed commit: "feat(tts): add Volcengine (ByteDance) sp..." | Re-trigger Greptile}

Note on voice catalog: The current listVoices implementation returns 6 hardcoded popular Chinese voices. Volcengine actually offers 100+ voices across multiple languages and styles (see voice catalog). Happy to add dynamic voice listing via the Volcengine API in a follow-up if that would be useful.

@steipete Would love to get your review on this when you have a moment. This adds Volcengine (ByteDance) as a TTS speech provider — it's the major speech API in China and provides high-quality Chinese neural voices. We're using it in production with OpenClaw on Feishu (Lark) and it works well for native voice message delivery. All CI checks are passing. Thanks!

Rebased/reworked this PR onto current main and pushed maintainer edits to the PR branch.

What changed:

• Ported the Volcengine speech provider onto the current plugin/manifest shape.
• Swapped the raw https client for the repo SSRF-guarded fetch helper.
• Added payload/auth/error coverage for the Volcengine TTS request shape, including native ogg_opus voice-note output.
• Added a gated live test at extensions/volcengine/tts.live.test.ts.
• Updated TTS/provider docs, i18n glossary, and changelog.

Local proof on pushed head 6eab06e3aa:

• pnpm test extensions/volcengine/tts.test.ts
• OPENCLAW_LIVE_TEST=1 pnpm test:live extensions/volcengine/tts.live.test.ts (skips cleanly without credentials)
• pnpm check:changed
• pnpm check:docs
• pnpm build

I still need a matching Volcengine Speech AppID + Access Token to do the real live API call. The ModelArk/OpenAI-compatible key is not enough for this endpoint. I tried the AppID Peter had available with the local token candidates and Volcengine returned 401 / load grant: requested grant not found in SaaS storage, which means the AppID/token pair is not valid for Speech TTS or the TTS grant is not enabled.

Please provide a limited test VOLCENGINE_TTS_APPID and VOLCENGINE_TTS_TOKEN pair, or confirm a Speech Console app/token that can synthesize TTS, and I can run the live test before landing.

Maintainer follow-up pushed in 0be34792b1.

What changed:

• Reworked TTS auth to support the current BytePlus Seed Speech API-key flow (VOLCENGINE_TTS_API_KEY / BYTEPLUS_SEED_SPEECH_API_KEY) via https://voice.ap-southeast-1.bytepluses.com/api/v3/tts/unidirectional.
• Kept the legacy Volcengine AppID/token path for older Speech Console applications.
• Made resourceId, appKey, baseUrl, and voice configurable.
• Updated the default resource to seed-tts-1.0: the default BytePlus project key live-tested here is valid, but seed-tts-2.0 returns requested resource not granted unless the project has TTS 2.0 entitlement.
• Removed generic BYTEPLUS_API_KEY fallback from TTS detection because it can be a ModelArk key and caused a false-positive auth path here.

Validation:

• pnpm test extensions/volcengine/tts.test.ts
• OPENCLAW_LIVE_TEST=1 pnpm test:live extensions/volcengine/tts.live.test.ts with .profile Seed Speech key; returned real MP3 bytes (ID3 header)
• pnpm check:docs
• pnpm check:changed
• pnpm build

Dependency/source check: verified against BytePlus Seed Speech unidirectional HTTP docs and voice list; the live API behavior matches the API-key header flow and current default-project entitlement.

Rewrote this PR branch history to one clean commit: 115f0cf794 feat(tts): add BytePlus Seed Speech provider.

The tree is identical to the previously validated head (0be34792b1); this was history cleanup only. The squashed commit keeps Rui as author and Peter as co-author.

Landed via GitHub rebase onto main.

• Source SHA: 115f0cf
• Landed SHA: 1531123
• Gates: pnpm test extensions/volcengine/tts.test.ts; OPENCLAW_LIVE_TEST=1 pnpm test:live extensions/volcengine/tts.live.test.ts; live .profile Ogg/Opus probe (OggS bytes); pnpm check:changed; pnpm build; exact-head PR CI green.
• API check: verified against BytePlus Seed Speech docs for the unidirectional HTTP API and voice list.

Thanks @xuruiray!