app.py

from fastapi import FastAPI, Request, HTTPException
from transformers import pipeline
import io
import librosa
from transformers import WhisperForConditionalGeneration, WhisperProcessor

app = FastAPI()
# Device configuration
# Load the model and processor
model_id = "WajeehAzeemX/whisper-smal-ar-testing-kale-5000"
model = WhisperForConditionalGeneration.from_pretrained(
    model_id
)
import torch

processor = WhisperProcessor.from_pretrained('WajeehAzeemX/whisper-smal-ar-testing-kale-5000')
forced_decoder_ids = processor.get_decoder_prompt_ids(language="arabic", task="transcribe")

from transformers import GenerationConfig, WhisperForConditionalGeneration
generation_config = GenerationConfig.from_pretrained("openai/whisper-small") # if you are using a multilingual model
model.generation_config = generation_config

pipe = pipeline(
    "automatic-speech-recognition",
    model=model,
    tokenizer=processor.tokenizer,
    feature_extractor=processor.feature_extractor,
    
)

@app.post("/transcribe/")
async def transcribe_audio(request: Request):
    try:
        # Read binary data from the request
        audio_data = await request.body()

        # Convert binary data to a file-like object
        audio_file = io.BytesIO(audio_data)
        
        # Load the audio file using pydub
        audio_array, sampling_rate = librosa.load(audio_file, sr=16000)

        # Process the audio array
        input_features = processor(audio_array, sampling_rate=sampling_rate, return_tensors="pt").input_features 

        # Generate token ids
        predicted_ids = model.generate(input_features, forced_decoder_ids=forced_decoder_ids)

        # Decode token ids to text
        transcription = processor.batch_decode(predicted_ids, skip_special_tokens=True)

        # Print the transcription
        print(transcription[0])  # Display the transcriptiontry:
       
        return {"transcription": transcription[0]}
    except Exception as e:
        raise HTTPException(status_code=500, detail=str(e))