Transcribe Service

Version: 1.0.8

Drop video or audio here
or click to choose a file

No file selected.


  

Supported formats: Any video or audio format supported by ffmpeg — including MP4, MOV, MKV, WebM, AVI, MPEG, MP3, WAV, M4A, AAC, OGG, FLAC, Opus, WMA.

Memory

6,098 MBUsed
9,469 MBAvailable
15,568 MBTotal
0Active jobs

Jobs

Loading…

No active jobs.

POST /transcribe

Returns text/event-stream. JSON body with gcs_url or multipart upload at POST /transcribe/upload.

{
  "gcs_url": "https://storage.googleapis.com/bucket/video.mp4",
  "model": "small",
  "hotwords": ["AdminPulse", "AccountancyVandaag"]
}

Models

Pass model to select a whisper model. Multilingual models auto-detect language; language-specific entries load fine-tuned weights where available or pin a multilingual model to a language (large-v3.nl, large-v3.fr). Legacy quality presets still work (fast → small, balanced → medium, best → large-v3-turbo).

modellanguagesweightscomputeRAM / job
smallmultilingualsmallint8~2 GB
mediummultilingualmediumint8~4 GB
large-v3-turbomultilinguallarge-v3-turboint8~7 GB
distil-large-v3multilingualdistil-large-v3int8~9 GB
large-v3multilinguallarge-v3int8~14 GB
small.enensmall.enint8~2 GB
medium.enenmedium.enint8~3 GB
distil-small.enendistil-small.enint8~2 GB
distil-medium.enendistil-medium.enint8~3 GB
small.nlnlammaraldirawi/faster-whisper-small-nl-int8int8~2 GB
medium.nlnlmediumint8~4 GB
large-v3.nlnllarge-v3int8~14 GB
small.frfrartkeep/faster-whisper-small-cv11-frenchint8~2 GB
medium.frfrm4th1eu/faster-whisper-medium-french-ct2int8~4 GB
large-v3-french-distilfrbrandenkmurray/faster-whisper-large-v3-french-distil-dec16int8~8 GB
large-v3.frfrlarge-v3int8~14 GB

SSE events

event: started
event: log
event: segment
event: done
event: error

Supported file types

Any video or audio format supported by ffmpeg — including MP4, MOV, MKV, WebM, AVI, MPEG, MP3, WAV, M4A, AAC, OGG, FLAC, Opus, WMA.

ffmpeg decodes the file server-side; the browser accepts any video/* or audio/* MIME type.

Auth

Optional — currently open (no AUTH_TOKEN set).

GET /health

JSON status, loaded models, active jobs, memory stats.

GET /status

Live memory + per-job params, progress, and recent log messages (polled by this page).

POST /jobs/{id}/cancel

Request cancellation of an active job. Queued jobs waiting on the inference lock abort within ~5s; running jobs stop between whisper segments. In-flight ffmpeg may finish first.