Svara Mirror
Sing or hum into your microphone and watch your melody draw itself as a live pitch trail while Svara Mirror finds your tonic on its own and names the raga or scale you are singing — in Carnatic or Western terms. When two candidates are close, it tells you which single note settles it. A Song mode estimates the key of music playing nearby. Everything runs on your device: nothing is recorded, stored, or uploaded.
How to use Svara Mirror
- Tap "Enable microphone" and allow access — or tap "Watch a demo" to see the mirror run on a simulated voice first.
- Start by holding your Sa: one steady note for a couple of seconds anchors the mirror to your voice.
- Sing or hum slow phrases, holding notes at least half a second. Your melody draws itself as a glowing pitch trail, and the big readout names each note with a cents gauge.
- Keep singing — the mirror finds your Sa (tonic) automatically and ranks the ragas or scales that match, sharpening as it hears more of you. Or pin your Sa with the Fixed toggle.
- Toggle the built-in tanpura for a soft drone at your Sa — the classic way to hold your shruti while you sing.
- When two candidates are close, follow the hint: it names the one note that settles the tie.
- Switch to Song mode to estimate the key of music playing near the mic, or toggle Carnatic/Western to change the vocabulary.
Frequently asked questions
- Is my voice recorded or uploaded?
- No. Audio is analysed in real time on your own device and immediately discarded. Nothing is recorded, stored, or sent to any server — close the tab and it is gone.
- How does it find my Sa without being told?
- It weighs every note you sing by how long you dwell on it, with extra weight for the notes you rest phrases on — singers naturally return to their tonic. Once one pitch clearly leads, that becomes your Sa, shown with its frequency in Hz.
- How accurate is it?
- For solo singing or humming in a reasonably quiet room, pitch tracking is accurate to a few cents and note identification typically exceeds 95% — browser noise processing is disabled, low-confidence frames are gated, and glides are kept out of the analysis. Accuracy drops with heavy background noise or Bluetooth headset mics.
- Can it identify the scale of a song playing on speakers?
- Song mode estimates the key from a pitch-class profile of the audio. Polyphonic music is a fundamentally harder read than a single voice, so it is framed as a strong guess rather than a verdict.
- Why is the detected scale a different letter than the song’s original key?
- Because the mirror names the scale in YOUR key. Nearly everyone transposes songs into their comfortable range — sing an E♭-major song from D and the scale family is identical, only the letter moves. Set the optional “Original key” in the panel and the mirror shows exactly how far you’ve transposed.
- What if my singing isn’t pitch-perfect?
- The mirror tunes itself to you, not the piano: it tracks your personal reference as it drifts and names notes relative to it, so consistently-flat or sharp singing still reads as the scale you meant. When one note runs persistently off, it tells you which one and by how much.