Skip to main content
A
Docs

Dialect Detection

Classify Arabic speech into 5 dialect families — Gulf, Egyptian, Levantine, Maghreb, and MSA.

Arabic Dialect Families

Arabic is not a single language — it spans dozens of regional dialects that differ significantly in vocabulary, phonology, and grammar. Voice classifies Arabic speech into 5 broad dialect families, enabling government systems to route, tag, and analyze calls by regional origin.

FamilyRegionsCharacteristics
Gulf (Khaliji)UAE, Saudi, Qatar, Bahrain, Kuwait, OmanDistinctive pronunciation of qaf, Persian loanwords
EgyptianEgyptMost widely understood dialect, media influence
LevantineSyria, Lebanon, Jordan, PalestineSofter consonants, unique verb conjugations
MaghrebMorocco, Algeria, Tunisia, LibyaFrench/Berber influences, compressed vowels
MSAPan-ArabFormal Modern Standard Arabic, news and official communications

How Detection Works

Dialect detection runs as a post-processing step after Whisper transcription. The pipeline:

  1. Whisper transcribes the audio into Arabic text
  2. Dialect classifier analyzes lexical markers, phonological patterns, and morphological features in the transcribed text
  3. Confidence score is assigned to each dialect family, with the highest-scoring family selected as primary

The classifier leverages allam-2-7b (SDAIA's Arabic LLM) for nuanced dialect understanding, combined with rule-based heuristics for high-confidence lexical markers unique to each dialect family.

Gulf Dialect Priority

Since Anar is purpose-built for GCC governments, the Gulf dialect classifier is tuned for higher precision on UAE, Saudi, Qatari, and Omani speech patterns, distinguishing sub-regional variations within the Gulf family.

Use Cases

Government Call Center Routing

Dialect detection enables automatic routing of incoming calls to agents who speak the caller's regional dialect. A Gulf Arabic speaker calling a UAE government service line gets connected to an agent fluent in Khaliji, rather than an MSA-only operator.

Demographic Analytics

Understanding the dialect distribution of citizen interactions helps government agencies plan multilingual service coverage. If 40% of calls to a particular office come in Egyptian Arabic, that office may need Egyptian-dialect-trained staff.

Document Language Tagging

When Voice transcriptions feed into downstream systems like Anar Docs or Anar Chat, dialect metadata enriches the content for better retrieval and response generation. A RAG system can prioritize Gulf Arabic sources when answering a Gulf Arabic query.

Integration with Transcription

Dialect detection is included in the standard transcription response. The language field returns the ISO language code, and when the language is Arabic, the dialect family is available in the transcription metadata.

curl -X POST http://localhost:8003/api/v1/transcribe \
  -F "file=@call_recording.mp3" \
  -G -d "language=ar"
{
  "id": "tx-042",
  "text": "السلام عليكم، أبغى أسأل عن المعاملة حقتي",
  "language": "ar",
  "duration": 3.8,
  "segments": [
    {"start": 0.0, "end": 1.2, "text": "السلام عليكم"},
    {"start": 1.2, "end": 3.8, "text": "أبغى أسأل عن المعاملة حقتي"}
  ],
  "model": "whisper-large-v3",
  "created_at": "2024-01-15T10:30:00Z"
}

In this example, lexical markers like "أبغى" (I want, Gulf) and "حقتي" (mine, Gulf) are strong signals for Gulf dialect classification.

Dashboard Analytics

The Voice analytics dashboard displays dialect distribution charts showing the breakdown of detected dialects across all transcriptions. This helps operations teams understand the linguistic profile of their citizen interactions and plan resources accordingly.

TranscriptionVoice Agents