feat: configurable STT language and optional LLM normalization

[kt315]

Summary

Two backwards-compatible improvements to the STT pipeline:

1. Configurable language for local whisper — new sttLanguage option.
2. LLM normalization is now optional — STT works with no endpoint/model.

1. sttLanguage (local whisper)

Local whisper-cli previously ran without an explicit language, defaulting to
English, so non-English speech was transcribed as English. The language is now
configurable via a new option:

["@renjfk/opencode-voice", { "sttLanguage": "auto" }]

• New option sttLanguage, default "auto" (auto-detection).
• Passed to local whisper-cli as -l <sttLanguage> (e.g. -l ru, -l auto).
• Does not affect the remote STT API path or runtime model selection.

2. Optional LLM normalization

Previously the plugin required an LLM endpoint/model, otherwise every
transcription warned or errored. Now:

• If endpoint and model are configured — normalization runs as before.
• If LLM config is absent — the raw Whisper transcription is used directly.
  This is a supported mode, not an error/fallback, and produces no warnings.
• Real LLM failures (endpoint configured but API down) keep the existing
  raw-input fallback.
• TTS behavior is unchanged (still reports unavailability if LLM is not configured).

The minimal config for STT-only is now just:

["@renjfk/opencode-voice", {}]

Changes

• lib/stt.js — buildWhisperCliArgs (adds -l), skip normalization when LLM unconfigured.
• lib/llm-client.js — createClient now returns isConfigured.
• index.js — thread isConfigured into registerSTT.
• README.md — document sttLanguage and optional LLM.
• Tests — buildWhisperCliArgs and isConfigured.

Verification

• npm test — 12 pass
• npm run check — clean

[Copilot]

Pull request overview

This PR makes the speech-to-text (STT) pipeline more flexible by (1) allowing the local whisper-cli language to be configured and (2) making LLM-based transcription normalization optional so STT can run with an empty plugin config.

Changes:

• Add sttLanguage option (default "auto") and pass it to local whisper-cli via -l.
• Make LLM normalization conditional on LLM configuration (endpoint + model), while keeping the raw-transcription fallback for real LLM failures.
• Extend createClient() to report isConfigured, thread that through plugin initialization, and add tests/docs for the new behavior.

Reviewed changes

Copilot reviewed 6 out of 6 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
lib/stt.js	Adds buildWhisperCliArgs() with -l, and skips normalization when LLM isn’t configured.
lib/llm-client.js	Returns isConfigured alongside complete.
index.js	Wires isConfigured into registerSTT so normalization becomes optional.
test/stt.test.js	Adds unit tests for buildWhisperCliArgs() language flag behavior.
test/llm-client.test.js	Adds unit test coverage for isConfigured behavior.
README.md	Documents sttLanguage and that STT works without LLM configuration.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.