52 languages. Deeper accuracy for South Asian music.
Language-specific AI layers handle the phonological nuances that general models miss — tested on real song vocabulary, not generic speech corpora.
Language layers, not just language codes.
Selecting a language in lyrcs does more than route to a different model. It activates a prompt layer built for that language's musical vocabulary — sandhi rules, honorifics, regional spelling variants, and script-specific post-processing — at the transcription step of the pipeline.
Every transcription ships with an English translation and cultural notes — idioms, religious references, traditional terms — formatted for sync licensing and publishing metadata.
Four languages have dedicated correction layers today — Punjabi, Hindi, Tamil and Telugu — with Marathi on its own prompt layer and a shared Indian-language layer beneath the rest.
Every job that can have one also gets a second opinion: an independent speech engine hears the same audio, and the lines where it disagrees with us are flagged for you at the review step — surfaced for a human, never rewritten silently. It is the one check in the pipeline that regularly catches a real mis-heard word. 44 of 52 languages have a second listener today; the ones that don't are marked below.
pahir (ਪਹਿਰ) is a traditional Punjabi unit of time — roughly three hours, historically used to divide the night into watches. Carries poetic weight; signals deep night, not merely after sunset. Relevant for sync licensing notes.
Organised by script and region.
Language-specific correction layer. Tested heavily on Bollywood and Punjabi pop catalogues.
Correction layer for Hindi and a prompt layer for Marathi. One script, many regional languages.
Language-specific correction layer. Tested on classical and contemporary Tamil film music.
Language-specific correction layer.
Covered by the shared Indian-language layer.
Added for gospel and Afrobeats catalogues. Note that many display fonts cannot spell Yoruba, Igbo or Hausa correctly — we pick ones that can.
Need a language not listed?
We add languages based on catalogue demand. Tell us what you're working with and we'll let you know what's possible.
“We support any language that can be transcribed from audio — which covers most of the world. The difference is depth: for the languages above, we've built specific correction and transliteration layers that meaningfully improve accuracy on music.”