Greptile Summary

This PR adds MiniMax T2A V2 as a fourth TTS provider. The implementation is largely well-structured — config types, Zod validation, directive parsing, and the core minimaxTTS function are all correct. However, there is one critical logic bug: textToSpeechTelephony has no handler for provider === "minimax" and silently falls through to the OpenAI code path, sending the MiniMax API key to the OpenAI endpoint — telephony will fail for any user whose only configured TTS provider is MiniMax.

• Bug (critical): textToSpeechTelephony in src/tts/tts.ts falls through to the OpenAI code path when MiniMax is active, using the wrong credentials and endpoint.
• Doc/schema mismatch (minor): The vol JSDoc says 0–10 but the Zod schema uses .positive() (excludes 0); these should be aligned.

Confidence Score: 2/5

• Not safe to merge without fixing the telephony fallthrough bug that causes MiniMax to use the wrong API key and endpoint.
• The non-telephony path (textToSpeech) is correctly implemented and the config/validation work is solid, but textToSpeechTelephony has a clear logic error where MiniMax falls through to the OpenAI code path. This will silently fail or corrupt requests for telephony callers using MiniMax as their primary provider, which is a regression risk.
• src/tts/tts.ts — specifically the textToSpeechTelephony function which needs a MiniMax branch or explicit unsupported error.

This is a comment left during a code review.
Path: src/tts/tts.ts
Line: 836-883

Comment:
**MiniMax silently falls through to OpenAI in telephony**

`textToSpeechTelephony` has no `provider === "minimax"` branch. When MiniMax is the active provider, execution falls through to the OpenAI code path after the `elevenlabs` block. The MiniMax API key (resolved by `resolveTtsApiKey(config, "minimax")`) is then passed directly into `openaiTTS` with `config.openai.baseUrl` and `config.openai.model`. The OpenAI endpoint will reject the MiniMax API key, causing telephony to fail for any user whose only configured provider is MiniMax.

A dedicated MiniMax branch (similar to the one in `textToSpeech`) needs to be added here, or MiniMax should be explicitly pushed aside with an `"unsupported for telephony"` error like Edge TTS.

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

---

This is a comment left during a code review.
Path: src/config/types.tts.ts
Line: 81

Comment:
**`vol` doc says 0 is valid; Zod schema excludes it**

The JSDoc comment says `Volume (0–10, default 1.0)`, implying `0` is a valid lower bound. However, the Zod schema uses `z.number().positive()` which means `> 0` (strictly), so `0` would be rejected at validation time.

Either the comment should be updated to reflect that the minimum is `> 0` (e.g., `(0–10]`), or the Zod schema should use `z.number().min(0)` if `0` is intentionally supported.

```suggestion
    /** Volume (>0–10, default 1.0). Values must be strictly positive. */
    vol?: number;
```

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

_{Last reviewed commit: 307f981}

Closing as superseded by the landed MiniMax speech-provider implementation on current main.

MiniMax TTS is already in-tree through the bundled MiniMax plugin path:

• extensions/minimax/speech-provider.ts
• extensions/minimax/tts.ts
• extensions/minimax/index.ts
• docs/tools/tts.md

That landed on the extension-owned/provider-owned surface rather than the older core src/tts/* path proposed here, so this PR no longer makes sense to merge in parallel.

Thanks for putting this together.