VoiceLab is a voice quality analysis tool that runs entirely in your browser. It measures the things you can check before you publish a voice recording: signal-to-noise ratio (how far the voice sits above the noise floor), sibilance energy (the harsh s and t sounds that fatigue listeners), pause energy (what is left in the pauses after speech: noise or reverb, not told apart), speech ratio and speaking rate, long silences, and clipping. Each metric comes back as a number, not a feeling, so two takes can be compared honestly.

That makes it a practical voice debugging tool: when a podcast take sounds off, the numbers say whether the problem is the gain staging (clipping, noise floor), the mic distance (SNR), or the delivery (speaking rate, silences). They do not tell you how well a speech recogniser will cope: in our own test of 1,248 transcriptions, signal-to-noise did not predict recognition errors. They do work for comparing a TTS render with a human recording number by number, and for checking meeting or call recordings before you rely on them.

Everything runs on-device via the WebAudio API: your audio is analyzed locally and never uploaded. When you need the same checks server-side or in an AI agent, the hosted API and MCP server expose this exact analysis as the analyze_voice tool, so a pipeline or agent can check every render with the same numbers you see here.

Singing is a different question, so it has its own instrument below. It follows your pitch over time on a note grid and marks the stretches that stayed more than 25 cents out of tune for longer than a moment, which you can click to hear back. It also reports your vibrato rate and depth, and the range you actually held steady. One honest limit: without the melody it can tell you whether a note sits cleanly on a pitch, not whether it was the right note for the song. That distinction covers most of what singers mean by out of tune: drifting, scooping short, sitting between two notes.