Voice turns recorded calls into text, labels dialect and sentiment, and keeps each transcript searchable.
A recording cannot be searched for a name, a topic or a complaint.
Callers speak Gulf, Egyptian, Levantine and other forms of Arabic.
Supervisors cannot listen to every call.
Upload a file in one of 9 audio formats, up to 25 MB.
Sends the audio to a speech service. Then it labels dialect and sentiment on the text.
Voice stores it. Search it and see it in call analytics.
Anar supplies no models. You supply the key for the speech service.
Arabic, English or automatic detection. Nine audio formats, up to 25 MB for each file.
Voice keeps each transcript. Search across all of them.
Add names and terms that the speech service must know.
Five classes: Gulf, Egyptian, Levantine, Maghrebi and Modern Standard Arabic. Voice finds marker words in the text.
Voice splits the text into turns at pauses and questions. It does not identify voices.
A word list gives each call a sentiment. Call analytics shows the counts.
No accuracy figure exists yet for transcripts or labels.
Six voices with English names. It needs its own key. Arabic quality is not tested.
Audio in, a text answer out. It does not speak the answer.
A WebSocket takes audio chunks. Each chunk gets a full transcription.
Connect Voice to your phone system.
| Capability | Status | Note |
|---|---|---|
| Transcription, history, search | Shipped | Recorded files. Not real time. |
| Dialect, speaker turns, sentiment | Partial | Rules on the text. No accuracy figure. |
| Speech service endpoint | Partial | Fixed in the code. Not through Anar Gateway. |
| Speech output, voice assistant | Partial | The assistant answers in text only. |
| Access control | Partial | One customer per deployment. |
| Telephony, personal-data masking, consent | Roadmap | Planned. Not built. |
- Transcription of recorded calls
- History and search
- Call analytics
- Better speaker and dialect detection
- Arabic text preparation for speech output
- Arabic voices
- Custom vocabulary
- Mandatory personal-data masking
- A processing record for each request
- Real-time streaming and telephony
- Consent, retention and erasure
- A published GCC dialect accuracy test
The voice assistant can send its model call through the gateway. Transcription and speech output do not use it yet.
Voice can send its events to the Guard audit trail when you set the Guard address.
A one-month pilot: we deploy in your environment and run it with one team.
anarlabs.ai/contact