Baarcha · Fennec

Tunisian speech to text (ASR / STT) — Fennec

Fennec is Baarcha’s Tunisian automatic speech recognition (ASR) model. It turns spoken Derja into text—a task also called speech to text (STT). Record through your microphone or upload an audio file in the playground, including examples that mix Tunisian Arabic and French. Use the API to add transcription to your own application.

Try Fennec ↗API documentation Pricing and limits

What you can try

Turn an authorized voice note into a draft transcript you can search or summarize.
Evaluate after-call transcription using representative customer-service recordings.
Compare names, amounts and language switches against a human transcript.

Try it on your own example

Record a short fictitious message with one name, one number and a French phrase.
Write a reference transcript before running the model.
Upload the audio in the speech-to-text playground; its current file limit is shown there.
Compare the output with your reference, then try the multipart /api/v1/audio/transcriptions endpoint with model fennec.

A starting example

عسلامة، نحب نبدّل موعدي للجمعة مع العشرة متاع الصباح. ابعثلي confirmation بالـSMS.

An example to test, not a guaranteed response.

What to check

Recognition varies with background noise, speakers and recording quality. Do not infer production accuracy from a selected demo. Speaker separation, telephony integration and human review are separate parts of a call-center project. Use recordings you are allowed to process.

Evaluation worksheet ↗

Tunisian speech-to-text API

Create an API key in the dashboard and set BAARCHA_API_KEY on your server. Replace recording.mp3 with your audio file. Send the file and model as multipart form data; the default JSON response contains the transcript in its text field.

curl https://baarcha.tn/api/v1/audio/transcriptions \
  -H "Authorization: Bearer $BAARCHA_API_KEY" \
  -F file=@'recording.mp3' \
  -F model=fennec

API documentation ↗

Tunisian ASR and STT: common questions

What is the difference between Tunisian ASR and STT?

ASR means automatic speech recognition. STT means speech to text. Here, both describe converting spoken Tunisian Derja into written text. Fennec performs that transcription task; it does not generate a spoken reply.

Can Fennec transcribe Derja mixed with French?

Fennec supports Tunisian speech and French code-switching. Try recordings that resemble your actual use case and compare them with a human transcript. Pay particular attention to language changes, proper names, amounts and background noise.

How do I turn a Tunisian voice note into text?

Open the speech-to-text playground and upload your recording, or record through the microphone. Run the transcription and review the returned text. Developers can send the same kind of audio file to /api/v1/audio/transcriptions with model fennec.

Does transcription also translate or summarize the recording?

Transcription writes down the spoken content. Translation into another language and summarization are separate steps. You can pass a reviewed transcript to Hannibal with a translation or summary instruction, then check that the result preserves the meaning.