You can deploy this model and start earning money today!

bella-top/
bella-whisper-large-v3

Name: bella-whisper-large-v3
Availability: InStock

safetensors

whisper

audio

automatic-speech-recognition

openai/whisper-large-v3

finetune

mit

Views

No views yet

bella-whisper

bella-whisper is a series of variant models based on OpenAI Whisper, designed for accurate speech recognition transcription. By fine-tuning with thousands of hours of high-quality data, bella-whisper performs excellently in multiple benchmark tests, especially in the real estate agent domain.

Model List

Model Name	Base Model	Training Time	Hugging Face Repository
bella-whisper-large-v3-turbo	whisper-large-v3-turbo	Not Open	N/A
bella-whisper-large-v3	whisper-large-v3	20241118	Link
bella-whisper-medium	whisper-medium	Not Open	N/A
bella-whisper-small	whisper-small	Not Open	N/A
bella-whisper-tiny	whisper-tiny	Not Open	N/A

Highlights

🚀 Stable Performance: Stable performance in multiple general benchmark tests.
✏️ Punctuation Optimization: More accurate context-based punctuation completion.
💡 Specific Scenario Optimization: Specially optimized for the real estate agent domain.

Quick Start

Follow these steps to set up and install the runtime environment. We provide the following inference methods for model installation and execution.

Create Python Environment

bash

conda create -n bella-whisper python=3.10
conda activate bella-whisper

Install dependencies

bash

pip install --upgrade pip
pip install --upgrade transformers accelerate torch librosa

Run Inference

python

import torch
import librosa
from transformers import AutoModelForSpeechSeq2Seq, AutoProcessor, pipeline

device = "cuda:0" if torch.cuda.is_available() else "cpu"
torch_dtype = torch.float16 if torch.cuda.is_available() else torch.float32

model_id = "bella-top/bella-whisper-large-v3"

# Load model and processor
model =AutoModelForSpeechSeq2Seq.from_pretrained(
    model_id, torch_dtype=torch_dtype, low_cpu_mem_usage=True, use_safetensors=True
)
model.to(device)

processor = AutoProcessor.from_pretrained(model_id)

# Create inference pipeline
pipe = pipeline(
    "automatic-speech-recognition",
    model=model,
    tokenizer=processor.tokenizer,
    feature_extractor=processor.feature_extractor,
    torch_dtype=torch_dtype,
    device=device,
)

# Load local audio file
audio_path = "sample1.wav"  # Replace with your audio file path
audio, sample_rate = librosa.load(audio_path, sr=16000)  # Ensure sampling rate is 16kHz

# Perform inference using the pipeline
result = pipe(audio)
print("Transcription:", result["text"])

More info ref to docs