Developer platform

Playground

Start a conversation

In Derja, French, or both.

Your message

Enter to send

First token: –
Total: –
Chunks: –

Request

curl https://punica.tn/api/v1/chat/completions \
  -N \
  -H "Authorization: Bearer $BAARCHA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "hannibal",
    "messages": [
      {
        "role": "user",
        "content": "شكونك إنتي بالضبط؟"
      }
    ],
    "temperature": 0.7,
    "max_tokens": 512,
    "stream": true
  }'

Runs against the public v1 API with your key. This page executes the same call keylessly as a demo.

Transcribe your audio

Record or upload speech in Tunisian Derja.

Audio input

or or drop it here

MP3, WAV, M4A and more, up to 8 MB.

Model

Latency: –

Request

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

Runs against the public v1 API with your key. This page executes the same call keylessly as a demo.

Turn text into speech

Write in Tunisian Derja, pick a voice, add an expression where it belongs, and listen.

0 / 5,000Speed1.0×mp3

Dido reads Tunisian Derja; tags like [laughter] are performed, not spoken.

🇹🇳 Tunisian Derja. The examples on the right follow its conventions.

Dialects, styles and expression tags

Dialect is sent as language, an ISO 639-3 id. Dido is one model trained on these six and conditioned per request; text with no language is read as Tunisian. A built-in voice that carries its own dialect model ignores it.

Style is one of the engine's instruct phrases. The voice default keeps each voice's tuned delivery; Natural clears it.

Expressions are bracketed tags the engine performs instead of reading: [laughter], [sigh], [surprise-ah], [surprise-oh], [question-en], [confirmation-en], [dissatisfaction-hnn]. Type [ in the box to pick one, or use the menu. Anything else in brackets is read aloud as text.

Try an example

🇹🇳 Tunisian sentences. Choose one to fill the text box.

Derja as Tunisians write it, French and English kept in Latin script where speech switches into them.

Use your own voice

Clone a voice from a recording you have permission to use.

Built-in voices need no account.

Punica owns the rights to the built-in voices. Cloning requires consent.

Recording and consent

Use a 10–15-second recording and confirm your permission to clone the voice. Manage or delete voices on the platform; deleting a voice also erases its reference audio upstream.

First byte: –
Total: –

Request

curl https://punica.tn/api/v1/audio/speech \
  -H "Authorization: Bearer $BAARCHA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "dido",
    "input": "الشارع مسكر اليوم على خاطر الأشغال، لازم نعملو دورة طويلة باش نوصلو للخدمة في الوقت.",
    "voice": "female",
    "language": "aeb",
    "response_format": "mp3"
  }' \
  --output speech.mp3

Runs against the public v1 API with your key. This page executes the same call keylessly as a demo.

Clone a voice with the API

Request

curl https://punica.tn/api/v1/voices \
  -H "Authorization: Bearer $BAARCHA_API_KEY" \
  -F name='صوتي' \
  -F consent=true \
  -F sample=@'take-one.wav'

Returns a vc_… id to use as voice in the speech request. Consent is required. See the voices reference for recording requirements and voice status.