Whisper-Large-v3-Turbo-OpenVINO-INT8 (Silver Series)

This repository contains the Silver Series optimized OpenVINO™ IR version of Whisper-Large-v3-Turbo, quantized to INT8 precision using NNCF. This model is the gold standard for high-accuracy, low-latency local speech-to-text.

---

🐍 Python Inference (Optimum-Intel)

To run this Silver Series audio engine locally:

from optimum.intel import OVModelForSpeechSeq2Seq
from transformers import AutoProcessor

model_id = "CelesteImperia/Whisper-Large-v3-Turbo-OpenVINO-INT8"
processor = AutoProcessor.from_pretrained(model_id)
model = OVModelForSpeechSeq2Seq.from_pretrained(model_id)

# Process local audio file
audio_input = "path/to/audio.mp3"
inputs = processor(audio_input, sampling_rate=16000, return_tensors="pt")
generated_ids = model.generate(inputs.input_features, max_new_tokens=128)
print(processor.batch_decode(generated_ids, skip_special_tokens=True)[0])

---

💻 For C# / .NET Users (OpenVINO.GenAI)

Whisper Turbo is optimized for the native WhisperPipeline in the OpenVINO.GenAI framework, enabling high-performance transcription in Windows applications.

using OpenVino.GenAI;

// 1. Initialize the Transcription Pipeline
var device = "CPU"; // Use "GPU" to leverage your hardware
using var pipe = new WhisperPipeline("path/to/whisper-int8-model", device);

// 2. Execute Transcription
var result = pipe.Generate("path/to/audio.wav");
Console.WriteLine(result.Texts[0]);

---

🏗️ Technical Details

• Optimization Tool: NNCF (Neural Network Compression Framework)
• Quantization: INT8 Symmetric (Per-channel)
• Series: Silver (High Fidelity)
• Workstation Validation: Dual-GPU (RTX 3090 + RTX A4000)

---

☕ Support the Forge

Maintaining the production line for high-fidelity audio models requires significant resources. If these tools power your industrial projects or accessibility tools, please consider supporting the Forge:

Platform	Support Link
Global & India	Support via Razorpay

Scan to support via UPI (India Only):

---

📜 License

This model is released under the Apache 2.0 License.

---

Connect with the architect: Abhishek Jaiswal on LinkedIn