AI Voice Studio
Read text aloud inside VS Code with OpenAI, MiniMax, or MiMo TTS — no
intermediate files, no separate apps. Driven from a sidebar panel and a single
keyboard shortcut.
Migrated from the Raycast AI Voice Studio
extension. Same providers, same voice catalog, rebuilt for VS Code's Activity
Bar with chunked synthesis, pause/resume, and per-chunk prefetch.
Screenshots: TODO — sidebar, status bar progress, MiMo style chips.
Features
- Three providers, one switcher. OpenAI (
gpt-4o-mini-tts, tts-1,
tts-1-hd), MiniMax (speech-2.8-hd/2.6-hd/02-hd, mainland & global
endpoints), MiMo Token Plan (mimo-v2.5-tts, mimo-v2-tts).
- Sidebar UI. Activity Bar entry with provider / model / voice / speed +
inline text box. Voices grouped by category for easy scan.
- Chunked synthesis with 1-ahead prefetch. Long input is split at sentence
boundaries. The next chunk is being synthesized while the current chunk
plays, so the gap between chunks is essentially network latency.
- Pause / Resume / Stop. Pause is local (audio element); Stop aborts the
active fetch, drains the queue, and frees the session.
- MiMo emotion + audio-event tags. Toggleable chip rows inject
(开心), (严肃) style prefixes plus （笑声，叹息） event prefixes
before each chunk.
- Speaking-instructions support for
gpt-4o-mini-tts, plus speed /
sample-rate / bitrate / language-boost knobs for MiniMax.
- Status bar indicator. Synth spinner → playing
i/n → idle, click to
focus the sidebar.
- Secrets in SecretStorage. API keys never touch settings.json.
Quick start
- Install the
.vsix (or from the Marketplace once published):
code --install-extension ai-voice-studio-0.4.0.vsix
- Open the AI Voice Studio entry in the Activity Bar.
- Run
AI Voice Studio: Set API Key… from the Command Palette and pick the
provider you want. Paste your key. (Or just press Read with no key set —
the side panel offers a "Set API Key" button inline.)
- Select voice and speed in the sidebar. Type or paste text into the box and
click ▶ Read, or select text in the editor and press ⌘⌥R /
Ctrl+Alt+R.
- ⏸ Pause / ▶ Resume with the same primary button. ⏹ Stop or
⌘⌥S / Ctrl+Alt+S cancels the in-flight session.
Providers
OpenAI
- Models:
gpt-4o-mini-tts (default), tts-1, tts-1-hd.
- Voices:
cedar, alloy, verse, marin, plus the legacy tts-1 set.
- Speaking instructions are honored on
gpt-4o-mini-tts only —
aiVoiceStudio.openai.instructions lets you bias tone, pacing, language.
- Custom base URL via
aiVoiceStudio.openai.baseUrl (any OpenAI-compatible
endpoint).
MiniMax
- Models:
speech-2.8-hd (default), speech-2.6-hd, speech-02-hd.
- 15 curated voice IDs (English news, Chinese radio host, anime archetypes,
multilingual presets).
- Region switch —
mainland → api.minimaxi.com, global →
api.minimax.io.
- Server-side
speed (0.5–2.0), sampleRate, bitrate, languageBoost.
MiMo
- Models:
mimo-v2.5-tts (default, Chinese & English voices), mimo-v2-tts
(legacy).
- Use a Token Plan key (
tp-…). Pay-as-you-go sk-… keys are rejected
early with a clear message.
- Style chips → opening tag prefix
(开心), (唱歌), etc.
- Sound chips → audio event prefix
（笑声，叹息）.
- Free-form
aiVoiceStudio.mimo.stylePrompt is sent as a leading user message
for tone shaping.
Commands
| Command |
Default keybinding |
AI Voice Studio: Read Selection (or Clipboard) |
⌘⌥R / Ctrl+Alt+R |
AI Voice Studio: Stop Reading |
⌘⌥S / Ctrl+Alt+S |
AI Voice Studio: Set API Key… |
— |
AI Voice Studio: Clear API Key… |
— |
AI Voice Studio: Focus Sidebar |
— |
Settings cheat sheet
| Key |
Default |
Notes |
aiVoiceStudio.provider |
openai |
openai | minimax | mimo |
aiVoiceStudio.playbackRate |
1 |
0.5–4.0; client-side, no re-synth |
aiVoiceStudio.chunkSize |
250 |
Max chars per chunk (80–2000) |
aiVoiceStudio.openai.model |
gpt-4o-mini-tts |
|
aiVoiceStudio.openai.voice |
cedar |
|
aiVoiceStudio.openai.instructions |
"" |
Only on gpt-4o-mini-tts |
aiVoiceStudio.minimax.region |
mainland |
|
aiVoiceStudio.minimax.speed |
1 |
0.5–2.0 |
aiVoiceStudio.mimo.openingStyleTags |
[] |
Driven by Style chips |
aiVoiceStudio.mimo.audioEventTags |
[] |
Driven by Sound chips |
(Full schema in package.json → contributes.configuration.)
How chunking works
Long input is split at Chinese / English sentence terminators
(。！？；…\.!?;), merged so each chunk fits under chunkSize, and overflow
inside one sentence falls back to soft-break punctuation (，、,). The
session pipelines synthesis: while chunk i plays, chunk i+1 is already
being fetched. Stop aborts the in-flight fetch via AbortController;
Pause only suspends the <audio> element so the prefetch keeps filling
the queue.
Troubleshooting
- "X API key not set" → click the inline Set API Key button or run the
command. Keys live in VS Code's SecretStorage.
- "Use a MiMo Token Plan key (tp-…), not a pay-as-you-go sk- key." → MiMo's
TTS endpoint only accepts Token Plan keys.
- "Invalid voice/model for …" → switching models can leave a voice that
isn't supported. Re-pick a voice; the dropdown filters to the active model.
License
MIT — see LICENSE.
| |
