letzwhisper
Log in Get started

Private GPU · audio to MIDI · no rush

Send audio.
Get MIDI.

Audio becomes MIDI on our private GPU. Your audio is not sent to OpenAI. Same inexpensive queue as transcription — no speed guarantee.

Get started See the plan

Are you looking for transcription? We offer it.

Current wait

about 14 minutes

Queue is clear. Typical file, one at a time. Estimate, not a promise.

piano.wav queued → .mid 
MThd  6 bytes
MTrk  piano · violin
notes  C4  E4  G4  …

How it works

01

Upload or POST

Drop a recording in the app, or send multipart audio to the MIDI API with your key.

02

It waits in line

Speech and MIDI share one GPU. We don’t promise when. The homepage and your file list show a live estimate.

03

Download .mid

Watch the page or poll GET for the .mid. There is no MIDI webhook. Audio is already gone.

You get a standard .mid file. Audio is deleted as soon as MIDI is ready. The file is deleted after 10 days.

Plans

Start free. Subscribe when you have more audio. Checkout shows the price in your currency. If you’re in a hurry, this is still the wrong product — no speed SLA.

Free

$0 forever

  • ✓ 3 files per day
  • ✓ Maximum 20 minutes of audio
  • ✓ Longer files are dropped immediately
  • ✓ No speed SLA — delivery eventually
  • ✓ .mid download — audio deleted when ready
Start free

Whisper 100h

100h / month

Priced in your currency at checkout.

  • ✓ 100 hours of audio every month
  • ✓ Unused hours stay valid for 12 months
  • ✓ Files up to 500 MB
  • ✓ Cancel anytime — unused hours stay, this month is not refunded
  • ✓ Or refund the last paid month immediately, no questions — no chargeback needed
Create account, then subscribe

API

Always async. POST returns 202. Same free and paid hours as transcription. Poll for the .mid — we will not hold your HTTP request open. No MIDI webhooks.

Create

curl -X POST https://whisper.byel.dev/api/v1/midis \
  -H "Authorization: Bearer lw_live_…" \
  -F audio=@piano.wav

# 202
{"id":"m_01J…","status":"queued"}

Download

GET /api/v1/midis/m_01J…
GET /api/v1/midis/m_01J….mid

# .mid returns 409 while processing

Privacy

MIDI runs on our private GPU. Nothing is sent to OpenAI. We don’t listen to your audio, train on it, sell it, or use it for anything else. We write a MIDI file and delete the audio.

  • ✓MIDI conversion runs on our private GPU. Nothing is sent to OpenAI.
  • ✓Audio is deleted immediately after MIDI is generated.
  • ✓The .mid file is deleted after 10 days.
  • ✓Same account and Stripe hours as transcription.