Shqip Create account

Albanian speech-to-text · API and editor

Albanian,
written word for word.

Upload a recording: a meeting, an interview, a podcast, a session. Get the text with a timecode on every word. Speak like in Kosovo or like in Albania; the text comes out in standard Albanian.

Try it free below 15 free minutes with an account

Albanian transcription at a very low price: from €0.012 a minute, no subscription. Pay only for the seconds you use.

This is the voice of Albanian, written word for word.

This is the voice of Albanian, written word for word.

Animated example. Timecodes are illustrative.

Free trial

Try it with your own file

No account, no card. We transcribe the first 60 seconds of a file, with a timecode on every word.

  • Up to 100 MB
  • MP3, WAV, M4A, MP4, OGG, FLAC
  • Up to 3 trials a day
  • The file is deleted after 24 hours

Drop an audio or video file here

or

How it works

From voice to text in three steps

  1. Upload

    Drop the file in the browser or send it through the API. MP3, WAV, M4A, MP4 and more, up to 2 GB and 4 hours.

  2. Fjalim writes it

    Our Albanian model turns speech into text. Every word gets its own time, to the millisecond.

  3. Edit, export

    Correct the text or the captions in the editor. Export TXT, SRT or VTT, or fetch JSON through the API.

Two editors

Text and captions, next to the sound

The transcript editor follows the audio word by word. The caption editor checks every cue against the usual rules: 42 characters a line, 2 lines, 17 characters a second.

Speaker 1

Mirëdita. Sot flasim për zgjedhjet.

Speaker 1

Çështja kryesore është buxheti i komunës.

▮ 00:01.04〰 uncertain word

No.StartEndCPSCue
100:00.3200:02.9014Mirëdita. Sot flasim për zgjedhjet.
200:03.0200:05.4019Çështja kryesore është buxheti i komunës.
300:05.5600:08.1013Fjalën e ka kryetari i kuvendit.

19 chars/s, over the 17 limit

For developers

An API without surprises

Upload, create, fetch the result. API keys, HMAC-signed webhooks, and JSON with a timecode on every word.

  • Bearer keys, shown once
  • Signed webhooks, retried for up to 24 hours
  • JSON, TXT, SRT and VTT from the same transcript

Read the docs → API reference →

Create a transcription

curl https://api.fjalim.com/v1/transcriptions \
  -H "Authorization: Bearer $FJALIM_KEY" \
  -H "Content-Type: application/json" \
  -d '{"audio_url":"https://example.com/seanca.mp3",
       "webhook_url":"https://example.com/hooks/fjalim"}'

# 202 Accepted
{ "id": "6c1f…", "status": "queued", "billed_seconds": null }

When it finishes, we call your URL

POST /hooks/fjalim
Fjalim-Signature: t=1762128000,v1=9f2c…e41a

{ "event": "transcription.completed", "data": { "id": "6c1f…", … } }

Accuracy, without hype

How accurate is it?

On our full held-out tests, with recordings that were not used in training, Fjalim gets 7.6% of words wrong on read speech, 12.0% on broadcast and 11.9% on parliament.

Methodology
  • Word error rate (WER) counts substituted, deleted and inserted words against a human reference. Lower is better.
  • The tests use recordings that the model did not see during training.
  • Your audio will differ. Noisy rooms, overlapping speakers and strong dialect raise the error rate. Try your own file with the free trial.
Word error rate (lower is better)
Read speech 7.6%
Broadcast 12.0%
Parliament 11.9%

Kosovo, Albania

Speak like in Kosovo or like in Albania; the text comes out in standard Albanian.

People speak the way they speak. The text you get is written in standard Albanian, without changing the words that were said.

  • kam shkukam shkuar
  • ka baka bërë
  • due me ardhdua të vij

Illustrative examples of everyday speech. Real output depends on the audio.

Privacy

Your audio stays yours

Audio and transcripts are stored in the EU

Transcription runs on servers in the EU. Website delivery, payments and email may also be processed outside the EU, under the providers’ standard contractual clauses.

90 days hot, one year in total

Audio stays fast for 90 days, then moves to cold storage and is deleted after 365 days.

Delete any time

Deleting removes the audio and every transcript version immediately.

No training without your permission

We do not train models on customer audio without your permission.

Read the privacy policy → Data retention →

FAQ

Before you ask

Which languages does Fjalim support?

Albanian only. You can speak like in Kosovo or like in Albania; the text comes out in standard Albanian.

How much does it cost?

From €0.012 a minute, no subscription, billed per started second. You get 15 free minutes after verifying your email, and you can add credit from €10. See the pricing page.

How accurate is it?

On our full held-out tests, with recordings that were not used in training, Fjalim gets 7.6% of words wrong on read speech, 12.0% on broadcast and 11.9% on parliament. Noisy or multi-speaker audio does worse. Try your own file.

Which formats do you accept?

MP3, WAV, M4A, MP4, OGG, FLAC and others that ffmpeg reads. Up to 2 GB and 4 hours per file. The free trial takes up to 100 MB.

Are there speaker labels?

Not yet, but they are coming soon. Meanwhile, in the editor you can name speakers yourself, for example “Speaker 1”.

Is my audio used for training?

No. We do not train models on customer audio without your permission.

Can I delete my files?

Yes, any time, from the app or through the API. Deleting removes the audio and every transcript version immediately.

Is there real-time transcription?

It is planned, at €0.016 per minute. Today Fjalim transcribes files.

Write the word down. Do not lose it.

Create an account and get 15 free minutes.

Create a free account See pricing