Workspace / Speech studioPrivate by default
LESS TYPING. MORE TALKING.

Your voice,
in your words.

A little space for Tamil, Tanglish, and English.
Speak naturally. We’ll take it from here.

01 / STUDIO

Let’s hear it.

Choose your output language, then turn on the mic.

BROWSER INFERENCE
READY WHEN YOU ARE00:00
A thought, an idea, a whole conversation.
Microphone access requested on start
Speech engine loads once, when you’re ready.
THE WORDS, WITHOUT THE WORK

Your transcript

Processed on this device0 words · Editable
A CLOSER LOOK

How close did we get?

02 / EVALUATE

Add what was actually said to measure the difference.

Same language and script as your transcript.
WORD ERROR RATE
—%

Add a reference to calculate

CHARACTER ERROR RATE
—%

Add a reference to calculate

A note on the numbers

Lower is closer to the reference. We normalize Unicode to NFC, ignore case and punctuation, and collapse whitespace. Tamil vowel signs are preserved. WER compares space-separated words; CER compares Unicode code points, including spaces. Errors are substitutions + deletions + insertions, divided by the reference length. Rates can exceed 100%. An empty reference has no defined score. These are scores for your current transcript, not a model benchmark.

∿

Nothing to upload. Nothing to give away.

Your microphone audio and transcripts stay in this browser. The speech engine runs locally on your CPU. First use downloads the engine; speed depends on your device. Chrome or Edge recommended.

LOCAL
FIRST.