VocalCode vs Claude Code's built-in /voice
Claude Code ships push-to-talk dictation for supported Claude.ai accounts. It is integrated, does not consume Claude messages or tokens, and for its own prompt inputs may be exactly the right tool. This page is written by VocalCode, so check the linked Anthropic documentation.
What Claude Code's /voice actually is
Run /voice in the Claude Code CLI and hold Space to record, or run
/voice tap to make Space toggle recording on and off. Your speech appears in the prompt as you speak. It also works in the VS Code extension, and
transcription is tuned for coding vocabulary: regex, OAuth, JSON, and
your project and branch names are fed in as hints. Transcription does not consume Claude messages or tokens;
access still depends on a supported Claude.ai account and its applicable plan.
Three details decide most of the comparison, and all three are in Anthropic's own documentation:
- The audio goes to Anthropic's servers. "Audio is not processed locally."
- It needs a Claude.ai account. Not available when Claude Code runs on an Anthropic API key, Amazon Bedrock, Google Cloud's Agent Platform or Microsoft Foundry — and organisations with HIPAA compliance enabled have it switched off entirely.
- It types into Claude Code inputs. The CLI, VS Code extension and Claude Code agent view are supported. It is not system-wide dictation for unrelated desktop apps.
Side by side
| Claude Code /voice | VocalCode | |
|---|---|---|
| Where the text lands | Claude Code inputs | Most desktop text fields; elevated and secure fields may block insertion |
| Where speech is recognised | Anthropic's servers | Your own CPU |
| Account required | Claude.ai account | None |
| Speech recognition offline | No | Yes, after model download |
| Remote / SSH / web | Not supported | Runs where your keyboard is |
| Mandarin Chinese | Chinese is not supported — falls back to English | Dedicated Paraformer path |
| Billing | Included for supported Claude.ai use; no message/token usage | $4.99 once, plus tax |
If you dictate Mandarin Chinese, read this first
Claude Code's dictation supports twenty languages. Japanese and Korean are among them. Chinese is not. And an unsupported language does not simply refuse — the documentation is explicit about what happens instead:
“If your language setting is not in the supported list, /voice
warns you on enable and falls back to English for dictation.”
Speech decoded with the wrong language can produce plausible but incorrect words instead of a clear failure. We encountered that failure in an early VocalCode build: it guessed the language from the OS locale, so Chinese speech on some English-locale machines was decoded as English. We removed that guess and made the model route a first-run choice. VocalCode does not switch model routes per utterance, and you can change the selected route later in Settings. Within the European route, Parakeet automatically detects among its 25 documented languages; the individual language labels are previews and do not constrain decoding.
Mandarin Chinese here runs on a dedicated bilingual Chinese–English Paraformer model. Its upstream conversion includes mixed Chinese–English material, but VocalCode does not publish a controlled mixed-language accuracy claim without release-binary product evidence.
What we do not cover, so you hear it here rather than after paying: Cantonese, Thai and Vietnamese are not available in VocalCode. Korean and Japanese use a separate on-device SenseVoice route; as with every language, test the full 30-day trial on your own microphone and vocabulary before buying.
Use Claude Code's /voice if…
…you dictate into Claude Code and nowhere else, you are signed in with a Claude.ai account, and cloud transcription is fine with you. It is built in, does not consume messages or tokens, and is tuned for that prompt box.
Use VocalCode if…
- You also talk to Cursor, Zed, a browser, Slack, a code-review comment or a terminal
- Your audio cannot leave the machine — or you simply would rather it did not
- You are on an API key, Bedrock, Google Cloud's Agent Platform or Foundry, where
/voiceis unavailable - You need speech recognition on a plane or another disconnected environment
- You dictate in Chinese
The honest summary
These are not really competitors. Anthropic built a microphone into their own prompt box; we built one for the rest of the operating system. If Claude Code is the only place you ever speak, save your $4.99 — and if it is not, the trial is 30 days and does not ask for a card.