---
title: "How to transcribe Arabic audio with an API: curl and Python"
description: "Transcribe Arabic audio with BroadNet Transcribe Arabic through an OpenAI-compatible endpoint. Formats, the language field, Python and curl examples, and the price."
url: https://broadrouter.com/guides/arabic-speech-to-text-api/
lang: en
updated: 2026-11-16
publisher: BroadRouter (a BroadNet site); API by BroadNet Inference, https://inference.broadnet.ai
---

# Arabic speech-to-text API, step by step

Updated 2026-11-16

Send a WAV or MP3 file to https://inference.broadnet.ai/v1/audio/transcriptions with model broadnet/cohere-transcribe-arabic and language=ar. The response is JSON with the transcript in text. Convert other formats such as WhatsApp .ogg to WAV first. The price is per audio minute, and BroadNet says no audio or transcript is retained.

BroadNet Transcribe Arabic costs $0.0042 per audio minute. BroadNet reports it at #2 of 37 (its September 2026 figure; the live board listed it #4 of 39 on 15 November 2026) on the [Open Universal Arabic ASR Leaderboard](https://huggingface.co/spaces/elmresearchcenter/open_universal_arabic_asr_leaderboard), and #1 among models whose licences permit serving through an API. For other languages, BroadNet Transcribe covers 14 languages at $0.002 a minute. See the [BroadNet Transcribe Arabic model page](https://broadrouter.com/models/cohere-transcribe-arabic/) and the [BroadNet Transcribe model page](https://broadrouter.com/models/cohere-transcribe/).

Other providers can cost less. On 2026-11-14, [Soniox’s pricing page](https://soniox.com/pricing) listed about USD 0.10 per hour for file (async) transcription and USD 0.12 per hour for real-time, and its language list includes Arabic. That is below BroadNet Transcribe Arabic’s $0.25 an hour, so test both on your own audio.

## With curl

```bash
curl https://inference.broadnet.ai/v1/audio/transcriptions \
  -H "Authorization: Bearer $BROADNET_API_KEY" \
  -F model=broadnet/cohere-transcribe-arabic \
  -F language=ar \
  -F file=@voice-note.wav
```

The response is the transcript as JSON:

```json
{"text": "مرحباً، لقد وصلت إلى البريد الصوتي، اترك رسالتك من فضلك"}
```

## With Python

```python
from openai import OpenAI
import os

client = OpenAI(base_url="https://inference.broadnet.ai/v1", api_key=os.environ["BROADNET_API_KEY"])

with open("voice-note.wav", "rb") as audio:
    result = client.audio.transcriptions.create(
        model="broadnet/cohere-transcribe-arabic",
        file=audio,
        language="ar",
    )
print(result.text)
```

## Always send the language

Without `language`, Arabic can come back **transliterated into Latin letters**. Send `language=ar` for Arabic and `language=en` for English.

## Which Arabic dialects does it handle?

BroadNet describes BroadNet Transcribe Arabic as covering Arabic, including dialects, and publishes three word error rates for it. These are BroadNet’s own lab measurements from its [model catalogue](https://www.broadnet.ai/inference/models):

| Test set | What it is | Word error rate |
| --- | --- | --- |
| FLEURS-ar | Clean read sentences | 3.8 |
| MASC-ar | Mixed real-world Arabic speech | 21.1 |
| BroadNet dialect set | Dialect speech | 23.4 |

BroadNet does not publish a breakdown by dialect, so nobody outside BroadNet can say how Gulf audio scores against Egyptian or Maghrebi. Other providers state coverage differently. Deepgram lists Nova-3 Arabic with regional codes, from `ar-AE`, `ar-SA` and `ar-EG` to `ar-MA` and `ar-IQ`. Speechmatics sells one “Global Arabic” pack that names Modern Standard Arabic, Gulf, Egyptian and Levantine speech, plus an `ar_en` pack for Arabic mixed with English. Soniox lists Arabic (`ar`) without dialect detail.

A listed code is not a measured accuracy. Before you choose, score each candidate on your own clips grouped by dialect:

-   **Gulf (Khaleeji):** Saudi, Emirati, Kuwaiti, Qatari speech, often with English product names.
-   **Egyptian:** the most widely understood dialect in media.
-   **Levantine:** Syrian, Lebanese, Jordanian and Palestinian speech.
-   **Iraqi and Maghrebi:** Moroccan, Algerian and Tunisian speech is usually the hardest case.
-   **[Modern Standard Arabic](https://broadrouter.com/glossary/msa/):** news and formal speech, the easy baseline.

The [Arabic dialect speech recognition guide](https://broadrouter.com/blog/arabic-dialect-speech-recognition/) explains why dialects score worse, and the [word error rate explainer](https://broadrouter.com/blog/arabic-asr-wer-explained/) shows how to score them.

## How much do Arabic speech-to-text APIs cost?

Every provider below lists Arabic. Cheapest first; prices are for pre-recorded files unless marked, and were checked on 2026-11-14:

| Provider and model | Per minute | Per hour | Arabic coverage stated |
| --- | --- | --- | --- |
| Soniox, async (file) | about USD 0.0017 | about USD 0.10 | Arabic (`ar`) |
| Soniox, real-time | about USD 0.002 | about USD 0.12 | Arabic (`ar`) |
| [BroadNet Transcribe](https://broadrouter.com/models/cohere-transcribe/) (14 languages) | $0.002 | $0.12 | Arabic among 14 languages |
| [BroadNet Transcribe Arabic](https://broadrouter.com/models/cohere-transcribe-arabic/) | $0.0042 | $0.25 | Arabic, including dialects (BroadNet) |
| Deepgram, Nova-3 monolingual | USD 0.0043 | USD 0.258 | `ar` plus regional codes |
| Speechmatics, Standard | USD 0.0075 | USD 0.45 | Global Arabic, `ar_en` |
| Speechmatics, Enhanced | USD 0.0125 | USD 0.75 | Global Arabic, `ar_en` |

Sources: [Soniox pricing](https://soniox.com/pricing) (billed per token; the hourly figures are Soniox’s approximations), [Deepgram pricing](https://deepgram.com/pricing) and [Speechmatics pricing](https://www.speechmatics.com/pricing). Soniox is cheaper than both BroadNet models, and BroadNet does not document real-time streaming. BroadNet’s case is the Arabic-specific model, its reported leaderboard rank and zero data retention. For more providers, see our open [speech-to-text API price index](https://broadrouter.com/data/speech-to-text-api-prices/), which lists batch and streaming prices and Arabic support for more than 20 products with a source and date for each (CSV and JSON). For add-on costs, see [speech-to-text API pricing compared](https://broadrouter.com/blog/speech-to-text-api-pricing-compared/); to price your own volume, use the [speech-to-text cost calculator](https://broadrouter.com/tools/speech-to-text-cost-calculator/).

## Limits

-   WAV or MP3, up to 26,214,400 bytes (25 MiB) per file. A larger file returns 413; split it first.
-   Pay-as-you-go keys allow 30 requests a minute and 4 at once.
-   Both transcription models work on the free starting credit of $5, so you can test your own recordings before paying.

## Troubleshooting

-   **Latin letters instead of Arabic script.** Add `language=ar`.
-   **400 or 422.** Check the field names (`model`, `language`, `file`) and that the file is real WAV or MP3. A renamed `.ogg` fails; convert it with ffmpeg.
-   **413.** The file is over 26,214,400 bytes. Convert to 16 kHz mono WAV or MP3, or split it into shorter pieces.
-   **401 on a paid account.** Check the key and the pay-as-you-go allowance. See [BroadNet error codes and retries](https://broadrouter.com/guides/errors-and-retries/).
-   **Weak accuracy on dialect or noisy calls.** BroadNet’s own dialect test set scores worse than clean read speech (see the FAQ). Test on a sample of your real audio and measure the [word error rate](https://broadrouter.com/glossary/word-error-rate/).

WhatsApp voice notes arrive as `.ogg`. [Convert and transcribe WhatsApp voice notes in one step](https://broadrouter.com/guides/transcribe-whatsapp-voice-notes/). If you need numbers, dates and names in noisy audio, BroadNet offers a fine-tuned Arabic model [on request](https://broadrouter.com/models/#on-request).

## Terms in this guide

[Speech-to-text](https://broadrouter.com/glossary/speech-to-text/)

Speech-to-text, also called automatic speech recognition (ASR), is technology that converts spoken audio into written text. A speech-to-text API accepts an audio file or stream and returns a transcript, usually priced per minute of audio. Accuracy is measured with word error rate and varies with accent, dialect and recording quality.

[Dialectal Arabic](https://broadrouter.com/glossary/dialectal-arabic/)

Dialectal Arabic is the everyday spoken Arabic of a region or city, such as Egyptian, Gulf, Levantine or Maghrebi Arabic. Dialects differ from Modern Standard Arabic in sounds, words and grammar, have no standard spelling, and are increasingly written online, which makes them harder for language models.

[Modern Standard Arabic (MSA)](https://broadrouter.com/glossary/msa/)

Modern Standard Arabic (MSA) is the standardized form of Arabic used in writing, news, education and formal speech across the Arab world. It is nobody's native everyday language: people speak regional dialects such as Egyptian or Gulf Arabic and switch to MSA in formal settings.

[Pay-as-you-go](https://broadrouter.com/glossary/pay-as-you-go/)

Pay-as-you-go is a pricing model in which you pay only for what you use, metered per unit such as tokens, requests or audio minutes, with no subscription or minimum commitment. Most LLM APIs bill this way, from a prepaid balance or a periodic invoice.

[Word error rate](https://broadrouter.com/glossary/word-error-rate/)

Word error rate (WER) is the standard accuracy measure for speech recognition: the number of word substitutions, deletions and insertions needed to turn a transcript into the correct reference, divided by the number of words in the reference. Lower is better.

[WAV](https://broadrouter.com/glossary/wav/)

WAV (Waveform Audio File Format, or WAVE) is an audio container built on Microsoft and IBM's RIFF format. It usually holds uncompressed linear PCM samples, so its size is predictable: sample rate times bytes per sample times channels, per second.

[Streaming](https://broadrouter.com/glossary/streaming/)

Streaming is a way of returning a language model response in small chunks while it is still being generated, instead of as one reply at the end. OpenAI-compatible APIs stream over server-sent events when the request sets stream to true.

[Concurrency](https://broadrouter.com/glossary/concurrency/)

Concurrency is the number of requests a client has in flight at the same time. LLM APIs often cap it per key or plan, separately from requests per minute, because each open request holds GPU capacity until its response has finished.

## Questions

### Does it understand Arabic dialects?

BroadNet reports a 23.4 word error rate on its own dialect test set, alongside 3.8 on FLEURS-ar and 21.1 on MASC-ar. Lower is better. BroadNet does not publish a score per dialect, so test Gulf, Egyptian, Levantine and Maghrebi recordings separately before you rely on it.

### Which Arabic speech-to-text API is cheapest?

Of the providers checked on 2026-11-14, Soniox lists the lowest rate, about USD 0.10 per hour for file transcription. BroadNet Transcribe Arabic costs $0.25 an hour. Deepgram Nova-3 and Speechmatics cost more. Price per hour matters less than word error rate on your own dialect audio.

### Which audio formats are accepted?

WAV (audio/wav) and MP3 (audio/mpeg). Convert other formats such as .ogg or .m4a with ffmpeg first.

### Why is my Arabic transcript written in Latin letters?

The request did not include the language field. Without it, Arabic speech can be transliterated into Latin script. Send language=ar with every Arabic request.

### Can I try Arabic transcription for free?

Yes. Both BroadNet transcription models run on the free starting credit that new accounts get, so you can test your own recordings before you top up.

### Should I use BroadNet Transcribe Arabic or BroadNet Transcribe for Arabic?

Both accept language=ar. BroadNet Transcribe Arabic is the Arabic model and the one BroadNet reports on the Arabic leaderboard. BroadNet Transcribe costs less per minute and covers 14 languages, so test both on your audio.

## Related guides

-   [Transcribe WhatsApp voice notes to text with an API](https://broadrouter.com/guides/transcribe-whatsapp-voice-notes/)
-   [Read Arabic invoices and receipts with a vision model](https://broadrouter.com/guides/read-arabic-invoices-with-vision/)
-   [BroadNet Inference error codes and when to retry](https://broadrouter.com/guides/errors-and-retries/)

Try it on your real workload. New accounts start with $5 of credit for Gemma and transcription. DeepSeek, GLM and Qwen3.8 need a top-up.

[Get started](https://inference.broadnet.ai/account/?utm_source=broadrouter&utm_medium=referral&utm_campaign=guides-arabic-speech-to-text-api&utm_content=page-cta) ·

**Sources**

-   [Soniox speech-to-text pricing (checked 2026-11-14)](https://soniox.com/pricing)
-   [BroadNet Inference API reference, transcription](https://inference.broadnet.ai/docs?utm_source=broadrouter&utm_medium=referral&utm_campaign=guides-arabic-speech-to-text-api&utm_content=sources#requests)
-   [BroadNet Inference pricing](https://inference.broadnet.ai/pricing?utm_source=broadrouter&utm_medium=referral&utm_campaign=guides-arabic-speech-to-text-api&utm_content=sources)
-   [BroadNet model catalogue, speech to text (lab measurements)](https://www.broadnet.ai/inference/models)
-   [Open Universal Arabic ASR Leaderboard](https://huggingface.co/spaces/elmresearchcenter/open_universal_arabic_asr_leaderboard)
-   [Soniox docs, supported languages (Arabic listed; checked 2026-11-14)](https://docs.soniox.com/stt/concepts/supported-languages)
-   [Deepgram pricing, Nova-3 pre-recorded (checked 2026-11-14)](https://deepgram.com/pricing)
-   [Deepgram models and languages, Nova-3 Arabic codes (checked 2026-11-14)](https://developers.deepgram.com/docs/models-languages-overview)
-   [Speechmatics pricing, pay-as-you-go batch (checked 2026-11-14)](https://www.speechmatics.com/pricing)
-   [Speechmatics languages, Global Arabic and Arabic-English (checked 2026-11-14)](https://docs.speechmatics.com/speech-to-text/languages)
