Uz Whisper Small Stt v2

--- language: - uz pipeline_tag: automatic-speech-recognition tags: - whisper - faster-whisper - ctranslate2 - speech-recognition --- # Uzbek Whisper Small STT v2 An Uzbek automatic speech recognition model based on Whisper Small, fine-tuned on Mozilla Common Voice 26 and converted to CTranslate2 FP16 for efficient inference with `faster-whisper`. ## Evaluation | Metric | Value | |---|---:| | Training epochs | 3 | | Optimization steps | 9,195 | | Test samples | 12,405 | | Test WER | 22.1379% | | Test loss | 0.2795 | Results were measured on the Uzbek test split of Mozilla Common Voice 26. Lower WER is better. ## Installation ```bash pip install faster-whisper ``` ## Usage ### NVIDIA GPU ```python from faster_whisper import WhisperModel model = WhisperModel( "maqsudxo1ja/uz-whisper-small-stt-v2", device="cuda", compute_type="float16", ) segments, info = model.transcribe( "audio.wav", language="uz", beam_size=5, ) text = "".join(segment.text for segment in segments).strip() print(text) ``` ### CPU ```python from faster_whisper import WhisperModel model = WhisperModel( "maqsudxo1ja/uz-whisper-small-stt-v2", device="cpu", compute_type="int8", ) ``` ## Model Format The `main` branch contains CTranslate2 model files and should be loaded with `faster-whisper`. It is not a standard Transformers checkpoint. The previous Transformers-format revision is preserved in the [`backup-transformers-before-finetune-20260810`](https://huggingface.co/maqsudxo1ja/uz-whisper-small-stt-v2/tree/backup-transformers-before-finetune-20260810) branch. ## Limitations Performance may vary across microphones, background noise, speaking styles, dialects, and audio domains that differ from Mozilla Common Voice. Validate the model on representative production audio before deployment.

--- language: - uz license: apache-2.0 library_name: transformers pipeline_tag: automatic-speech-recognition base_model: maqsudxo1ja/uz-whisper-small-stt tags: - whisper - speech-recognition - uzbek - stt - asr - generated_from_trainer datasets: - islomov/it_youtube_uzbek_speech_dataset metrics: - wer model-index: - name: uz-whisper-small-stt-v2 results: - task: type: automatic-speech-recognition name: Automatic Speech Recognition dataset: type: islomov/it_youtube_uzbek_speech_dataset name: IT YouTube Uzbek Speech Dataset metrics: - type: wer value: 28.64 name: Word Error Rate --- # 🇺🇿 UZ Whisper Small STT v2 O'zbek tilidagi nutqni matnga aylantiruvchi (Speech-to-Text) model. `openai/whisper-small` arxitekturasi asosida, o'zbek tilidagi tabiiy suhbat uslubidagi nutqqa moslashtirilgan. Bu model [`maqsudxo1ja/uz-whisper-small-stt`](https://huggingface.co/maqsudxo1ja/uz-whisper-small-stt) modelining davomiy fine-tuning (continued fine-tuning) natijasi bo'lib, IT mavzusidagi YouTube videolaridan olingan real, so'zlashuv uslubidagi audio ma'lumotlar bilan yanada takomillashtirilgan. ## 📊 Natijalar | Ko'rsatkich | Qiymat | |---|---| | **WER (Word Error Rate)** | **28.64%** | | Validation Loss | 0.2634 | | Model hajmi | ~242M parametr (Whisper-small) | <details> <summary>To'liq training jarayoni (bosing)</summary> | Training Loss | Epoch | Step | Validation Loss | WER | |---|---|---|---|---| | 3.0847 | 0.68 | 400 | 0.3161 | 32.49 | | 2.4038 | 1.35 | 800 | 0.2827 | 29.58 | | 2.0104 | 2.03 | 1200 | 0.2708 | 30.53 | | 1.9056 | 2.70 | 1600 | 0.2647 | 28.92 | | 1.9695 | 3.00 | 1776 | 0.2634 | **28.64** | </details> ## 🚀 Tezkor boshlash ### O'rnatish ```bash pip install transformers torch librosa soundfile ``` ### 1-usul: `pipeline` bilan (eng oson) ```python from transformers import pipeline transcriber = pipeline( "automatic-speech-recognition", model="maqsudxo1ja/uz-whisper-small-stt-v2" ) result = transcriber("audio_fayl.mp3") print(result["text"]) ``` ### 2-usul: to'g'ridan-to'g'ri model bilan (batafsil nazorat) ```python import torch import librosa from transformers import WhisperProcessor, WhisperForConditionalGeneration MODEL_ID = "maqsudxo1ja/uz-whisper-small-stt-v2" processor = WhisperProcessor.from_pretrained(MODEL_ID, language="uzbek", task="transcribe") model = WhisperForConditionalGeneration.from_pretrained(MODEL_ID)...

Source context: 162 downloads · 3 likes · Pipeline automatic-speech-recognition · Library transformers · Repo maqsudxo1ja/uz-whisper-small-stt-v2