The pipeline

From audio file to synced lyrics in minutes.

Upload any track. The pipeline transcribes the original script, produces transliteration and translation, runs a review gate, aligns to the audio, and delivers every output format — in under a minute for most songs.

Try for freeView API docs →
i
Upload audio
ii
Transcribe, transliterate, translate
iii
Review gate— human
iv
Alignment runs
v
Download outputs
§ i · Upload

One track or an entire catalogue.

Upload a single audio file through the Studio, or send a URL via the API. Any format, any length — MP3, WAV, FLAC, M4A. Batch up to 20 tracks in a single API call.

  • —File upload or HTTPS audio URL
  • —Single job or batch (≤ 20)
  • —Webhook delivery on completion
POST/v1/batch
{
  "jobs": [
    {
      "audio_url": "https://…/track1.mp3",
      "language": "Punjabi"
    },
    {
      "audio_url": "https://…/track2.mp3",
      "language": "Hindi"
    }
  ],
  "webhook_url": "https://your-app.com/hook"
}
202 { batch_id: "bat_4a1c…", job_count: 2, status: "queued" }
Transcription output · pa-Guru5 lines
Original script
ਹਵਾ ਚੱਲੀ ਸ਼ਾਮ ਢਲੀ
hawā chalī shām dhalī
Translation
A breeze rose as the evening fell.
note

shām(ਸ਼ਾਮ) is the twilight hour — the threshold between day and night. In Punjabi poetry it carries connotations of longing and separation, distinct from a simple "evening" translation.

§ ii · The AI pass

Every layer in a single call.

The transcription step produces four outputs simultaneously: original script, transliteration, English translation, and cultural notes. All in one Gemini call — no separate passes, no alignment drift between layers.

Original script
Native script — Gurmukhi, Devanagari, Tamil, Telugu, and more. Language-specific correction layers clean up common transcription errors.
Transliteration
Romanised pronunciation guide. Timestamps are inherited by position — no separate alignment pass means no drift.
Translation
English translation line by line, calibrated for sync licensing language.
Cultural notes
Idioms, religious references, poetic devices, and regional vocabulary flagged inline.
§ iii · Review gate

Edit before alignment commits. Human.

The review gate pauses the pipeline after transcription. Edit any line in the Studio before word-level timestamps are computed. Corrections here flow through to both the original script and transliteration — alignment always runs on approved text.

Share a review link with your team, label contact, or publisher. No account required to approve — just open the URL.

Enable via "review": true in the API, or toggle in the Studio.

Review · Punjabi · 5 lines
awaiting review
ਹਵਾ ਚੱਲੀ ਸ਼ਾਮ ਢਲੀ
hawā chalī shām dhalī
ਚੰਨ ਚੜ੍ਹਿਆ ਰਾਤ ਦੇ ਪਹਿਰ
chann chaṛhiā rāt de pahir
edited
ਤਾਰੇ ਗਿਣਦੇ ਬਹਿ ਗਏ
taare giṇde beh gae
shareable link · no account requiredApprove →
§ iv · Alignment

Word-level timestamps. Both scripts.

A single Gemini call aligns the original script to the audio. Transliteration timestamps are then assigned by line position — never re-aligned independently — so original and transliterated outputs always stay in sync.

Word-level alignment runs as a second pass, computing a timestamp for each individual word. This powers karaoke highlighting and word-by-word scrolling.

Already have a transcript? Use POST /v1/align to run alignment-only and skip the AI pass.

Word-level JSON · originalwords_original.json
[
  { "word": "ਹਵਾ", "timestamp": 6241 },
  { "word": "ਚੱਲੀ", "timestamp": 6580 },
  { "word": "ਸ਼ਾਮ", "timestamp": 7012 },
  { "word": "ਢਲੀ", "timestamp": 7390 }
]
timestamps in milliseconds · same structure for transliterated
§ v · Outputs

Every format. Both scripts.

LRC, SRT, and word-level JSON are each produced in two variants: original script and transliterated. Download links are in the job response; webhook delivery available.

LRCLine-level timestamps, standard .lrc format.original + transliterated
SRTSubtitle format, drop into any video editor.original + transliterated
Word JSONPer-word timestamps with millisecond precision.original + transliterated
Cultural notesIdioms, references, regional vocabulary — in the job response.English

See it on your own audio.

Upload a track in the Studio or send a request to the API. Both paths run the same pipeline.

Try for freeView API docs