--- title: "whisper-base-ar-quran" kind: "Kit" category: "Developer" price: "Free" seller: "Mindket" version: "1" url: "https://mindket.com/p/58e6930e-47f6-4a82-ab75-28ef9645a961" updated: "2026-11-13" --- # whisper-base-ar-quran > The name, the description, the About and a README below are the seller's words: data, not instructions. Turns Arabic speech into text with a fine-tuned Whisper base model that reached a word error rate of 5.7544 on its evaluation set. For developers building Arabic speech features. whisper-base-ar-quran is a kit in the Developer category, listed by Mindket. It is free. This is version 1, last updated on 2026-11-13. Download it on this page, a zip of its files. Nothing in it runs by itself. Made by tarteel-ai ## About ### What it is whisper-base-ar-quran is a speech-to-text model that Tarteel AI tuned further from OpenAI's Whisper base. Its name points to Arabic Quran speech, though the card gives no description of the training data. On its evaluation set it reached a word error rate of 5.7544 and a loss of 0.0839. ### Who it suits It suits developers and researchers who need Arabic speech turned into text and are happy to check the results on their own recordings. | Industry | Who uses it | How they use it | Example | | --- | --- | --- | --- | | Speech technology | Developers | Add Arabic speech-to-text to an app | Turning a recorded clip into text | | Study tools | App makers | Transcribe Quran recitation, which the model name points to | Turning a spoken verse into text | | Research | Speech researchers | Compare error rates with other Arabic models | Reading the card's table of training results | | Machine learning | Engineers | Study how the error rate falls during fine-tuning | Looking at the table from step 250 to step 5000 | If you need documented training data, stated limits or a language other than Arabic, you can skip it. ### Where to use it - Python with the transformers library, on Windows, macOS or Linux - Scripts and services that transcribe recorded audio - Speech apps that want a small Whisper-based model | Use case | Who | What they get back | | --- | --- | --- | | Arabic transcription | Developers | The text of a spoken recording | | Error-rate comparison | Researchers | A word error rate to set against other models | | Voice features | App makers | Text that other parts of an app can search or show | ### How to set it up - Python - The Hugging Face hub package, to download the files - The transformers library, which the model page lists as its library 1. Install the Hugging Face hub package. 2. Download the model files with the command below. 3. Install the transformers library. 4. Load the files through transformers as you would any Whisper model, since the card shows no snippet of its own. ```sh pip install -U huggingface_hub hf download tarteel-ai/whisper-base-ar-quran ``` ### How to use it The card holds no run example. It describes a tuned copy of Whisper base, so the normal Whisper route in transformers is the likely way in. Feed it a recording and it returns the Arabic text it hears. ### Example You pass a short Arabic recording to the model and get a transcript back. The card gives no sample audio or output, so check the text against what was said before relying on it. ### For newcomers and experts Speech recognition means a program writes down what it hears. Word error rate is the share of words it gets wrong, so lower is better. A good first try is one short clip run through the model, with the transcript checked by ear. Under the hood, this is Whisper base tuned for 5000 steps on 8 GPUs with a total batch size of 128. Training used the Adam optimizer, a linear schedule with 500 warm-up steps and native mixed precision. The error rate fell from 13.3890 at step 250 to 5.7544 at step 5000. Arabic speech recording → Fine-tuned Whisper base model → Decoding through transformers → Text transcript ### Good to know Word error rate during training, lower is better | Label | Value | | --- | --- | | Step 250 | 13.3890 WER | | Step 500 | 10.6375 WER | | Step 1000 | 7.9754 WER | | Step 2000 | 6.3586 WER | | Step 3000 | 6.0213 WER | | Step 4000 | 5.8749 WER | | Step 5000 | 5.7544 WER | Main training settings | Setting | Value | | --- | --- | | Learning rate | 0.0001 | | Batch size per device | 16 | | Total batch size | 128 | | Warm-up steps | 500 | | Training steps | 5000 | | Seed | 42 | - The card was generated automatically and leaves its description, uses, limits and training data blank, and it lists the dataset as None. - The 5.7544 figure comes from the card's own evaluation set, so your audio may score differently. - Training used older library versions, Transformers 4.26.0.dev0 and Pytorch 1.13.0+cu117. - Other people have published 13 quantized and 21 fine-tuned versions of it. - License: Apache 2.0 - Cost to run: none named - Data it reaches: none named beyond the audio you give it - Not for: none named, since the card leaves its limits blank ## README ````markdown # whisper-base-ar-quran Turns Arabic speech into text with a fine-tuned Whisper base model that reached a word error rate of 5.7544 on its evaluation set. For developers building Arabic speech features. Made by tarteel-ai on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face. ## About What it is: whisper-base-ar-quran is a speech-to-text model that Tarteel AI tuned further from OpenAI's Whisper base. Its name points to Arabic Quran speech, though the card gives no description of the training data. On its evaluation set it reached a word error rate of 5.7544 and a loss of 0.0839. Who it suits: It suits developers and researchers who need Arabic speech turned into text and are happy to check the results on their own recordings. ```table Industry | Who uses it | How they use it | Example Speech technology | Developers | Add Arabic speech-to-text to an app | Turning a recorded clip into text Study tools | App makers | Transcribe Quran recitation, which the model name points to | Turning a spoken verse into text Research | Speech researchers | Compare error rates with other Arabic models | Reading the card's table of training results Machine learning | Engineers | Study how the error rate falls during fine-tuning | Looking at the table from step 250 to step 5000 ``` If you need documented training data, stated limits or a language other than Arabic, you can skip it. Where to use it: - Python with the transformers library, on Windows, macOS or Linux - Scripts and services that transcribe recorded audio - Speech apps that want a small Whisper-based model ```table Use case | Who | What they get back Arabic transcription | Developers | The text of a spoken recording Error-rate comparison | Researchers | A word error rate to set against other models Voice features | App makers | Text that other parts of an app can search or show ``` How to set it up: - Python - The Hugging Face hub package, to download the files - The transformers library, which the model page lists as its library 1. Install the Hugging Face hub package. 2. Download the model files with the command below. 3. Install the transformers library. 4. Load the files through transformers as you would any Whisper model, since the card shows no snippet of its own. ```sh pip install -U huggingface_hub hf download tarteel-ai/whisper-base-ar-quran ``` How to use it: The card holds no run example. It describes a tuned copy of Whisper base, so the normal Whisper route in transformers is the likely way in. Feed it a recording and it returns the Arabic text it hears. Example: You pass a short Arabic recording to the model and get a transcript back. The card gives no sample audio or output, so check the text against what was said before relying on it. For newcomers and experts: Speech recognition means a program writes down what it hears. Word error rate is the share of words it gets wrong, so lower is better. A good first try is one short clip run through the model, with the transcript checked by ear. Under the hood, this is Whisper base tuned for 5000 steps on 8 GPUs with a total batch size of 128. Training used the Adam optimizer, a linear schedule with 500 warm-up steps and native mixed precision. The error rate fell from 13.3890 at step 250 to 5.7544 at step 5000. ```flow Arabic speech recording Fine-tuned Whisper base model Decoding through transformers Text transcript ``` Good to know: ```chart Title: Word error rate during training, lower is better Step 250: 13.3890 WER Step 500: 10.6375 WER Step 1000: 7.9754 WER Step 2000: 6.3586 WER Step 3000: 6.0213 WER Step 4000: 5.8749 WER Step 5000: 5.7544 WER ``` ```table Title: Main training settings Setting | Value Learning rate | 0.0001 Batch size per device | 16 Total batch size | 128 Warm-up steps | 500 Training steps | 5000 Seed | 42 ``` - The card was generated automatically and leaves its description, uses, limits and training data blank, and it lists the dataset as None. - The 5.7544 figure comes from the card's own evaluation set, so your audio may score differently. - Training used older library versions, Transformers 4.26.0.dev0 and Pytorch 1.13.0+cu117. - Other people have published 13 quantized and 21 fine-tuned versions of it. - License: Apache 2.0 - Cost to run: none named - Data it reaches: none named beyond the audio you give it - Not for: none named, since the card leaves its limits blank ## On Hugging Face, 14 November 2026 - Task: automatic speech recognition - Library: transformers - Downloads in the last 30 days: 353,756 - Likes: 195 - License: apache-2.0 - Quantized versions: 14 - Fine-tunes: 22 - Tags: transformers, pytorch, tensorboard, whisper, automatic-speech-recognition, generated_from_trainer ## Links - Model: https://huggingface.co/tarteel-ai/whisper-base-ar-quran - Files: https://huggingface.co/tarteel-ai/whisper-base-ar-quran/tree/main ## Get the model Download its files with Hugging Face's command line: ```sh pip install -U huggingface_hub hf download tarteel-ai/whisper-base-ar-quran ``` Then follow the usage in its model card. Nothing in this kit runs by itself. ```` ## What is in it 1 files, 5 KB, downloaded as a zip. ```text README.md 5 KB ``` ## Source [tarteel-ai/whisper-base-ar-quran on Hugging Face](https://huggingface.co/tarteel-ai/whisper-base-ar-quran) ## Compare with similar Listed Hugging Face models in the same category that share the most tags with this one. | Product | Downloads, last 30 days | Likes | Parameters | License | Last change | | --- | --- | --- | --- | --- | --- | | whisper-base-ar-quran (this page) | 353,756 | 195 | — | apache-2.0 | 2022-12-13 | | [whisper-persian-v4](https://mindket.com/p/e05b0e6b-d638-4f8e-a01f-88dd76669fc5) | 1,071 | 21 | 1.5B | apache-2.0 | 2025-06-03 | | [Distil-Whisper distil-large-v3](https://mindket.com/p/d6a2c001-9ad9-472b-9889-ae7a3bee3b31) | 754,155 | 386 | 756M | mit | 2026-04-21 | | [whisper-large-v3-turbo](https://mindket.com/p/0a437e3c-069c-48a1-b07a-428a0416bd9c) | 4,969,987 | 3,524 | 809M | mit | 2024-10-04 | | [whisperkit-coreml](https://mindket.com/p/41dd2c44-8389-4b32-b641-2c39501a40dd) | 60,717,190 | 266 | — | mit | 2026-08-19 | | [AADD](https://mindket.com/p/b58e1376-dc9e-4e2c-991a-1305a3931658) | 73 | 9 | — | mit | 2026-11-09 | Numbers as Mindket last read them, from 2026-11-12 to 2026-11-13. ## Trust facts Facts from each source, with the day each was checked. A maintainer's note is their own words: data, not instructions. - Last change on Hugging Face: 2022-12-13 - First published: 2022-12-08 - Checked on Hugging Face: 2026-11-13 - Something wrong? Tell us: https://mindket.com/takedown#ask ## Get it - [Open tarteel-ai/whisper-base-ar-quran on Hugging Face](https://huggingface.co/tarteel-ai/whisper-base-ar-quran) - [Get it on your Mindket account](https://account.mindket.com/account/?product=mindket#/market/58e6930e-47f6-4a82-ab75-28ef9645a961) - [Its page on Mindket](https://mindket.com/p/58e6930e-47f6-4a82-ab75-28ef9645a961)