KamelTouati/whisper-small-darja-demo

--- language: - ar language_details: Algerian Arabic (Darja / الدارجة الجزائرية) license: mit tags: - whisper - audio - automatic-speech-recognition - speech-recognition - speech-to-text - peft - lora - darja - algerian-arabic - oddadmix - transformers - pytorch datasets: - oddadmix/arabic-audio-collection-algerian-kahwa-postcast - oddadmix/arabic-audio-collection-algerian-loubna-stories - oddadmix/arabic-audio-collection-algerian-rawi metrics: - wer base_model: openai/whisper-small pipeline_tag: automatic-speech-recognition model-index: - name: whisper-algerian-darja-small results: - task: type: automatic-speech-recognition name: Speech Recognition dataset: name: OddAdmix Algerian Loubna Stories type: oddadmix/arabic-audio-collection-algerian-loubna-stories metrics: - type: wer value: 14.87 name: Loubna Test WER (%) - task: type: automatic-speech-recognition name: Speech Recognition dataset: name: OddAdmix Algerian Rawi Stories type: oddadmix/arabic-audio-collection-algerian-rawi metrics: - type: wer value: 27.54 name: Rawi Test WER (%) - task: type: automatic-speech-recognition name: Speech Recognition dataset: name: OddAdmix Algerian Kahwa Podcast type: oddadmix/arabic-audio-collection-algerian-kahwa-postcast metrics: - type: wer value: 34.85 name: Kahwa Test WER (%) --- # Whisper Small — Algerian Arabic (Darja / الدارجة الجزائرية) --- ## Overview **`touati-kamel/whisper-algerian-darja-small`** is an Automatic Speech Recognition (ASR) model specifically fine-tuned for **Algerian Arabic (Darja / الدارجة الجزائرية)**. Built on top of **OpenAI's Whisper Small** (`openai/whisper-small`, 244M parameters), this model incorporates parameter-efficient LoRA adapters trained with **4-bit quantization (QLoRA)** over a sequential 3-phase curriculum covering conversational podcasts, spontaneous storytelling, and cultural narratives from the **OddAdmix Algerian speech collection**. ### Key Highlights - **Native Algerian Dialect Adaptation**: Handles authentic …