app.py

import streamlit as st
from st_audiorec import st_audiorec

from transformers import AutoModelForSpeechSeq2Seq, AutoProcessor, pipeline
#from datasets import load_dataset
import torch
from gliner import GLiNER

from resources import Lead_Labels, entity_labels, set_start, audit_elapsedtime

rec = None
ner = None
iteration = 0

def main ():

    print(f"Main iteration {iteration}")
    iteration += 1
    
    if rec is None:
        print("rec is None")
        rec = init_model_trans()
    if ner is None:
        print("ner is None")
        ner = init_model_ner() #async

    labels = entity_labels

    text = "I have a proposal from cgd where they want one outsystems junior developers and one senior for an estimate of three hundred euros a day, for six months."
    print(f"get entities from sample text: {text}")
    get_entity_labels(model=ner, text=text, labels=labels)

    print("Render UI")
    wav_audio_data = st_audiorec()

    if wav_audio_data is not None and rec is not None:
        print("Loading data...")
        st.audio(wav_audio_data, format='audio/wav')
        text = transcribe(wav_audio_data, rec)
        if text is not None:    
            get_entity_labels(labels=labels, model=ner, text=text)


def init_model_trans ():
    print("Initiating transcription model...")
    start = set_start()

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

    model_id = "openai/whisper-large-v3"

    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)

    pipe = pipeline(
        "automatic-speech-recognition",
        model=model,
        tokenizer=processor.tokenizer,
        feature_extractor=processor.feature_extractor,
        max_new_tokens=128,
        chunk_length_s=30,
        batch_size=16,
        return_timestamps=True,
        torch_dtype=torch_dtype,
        device=device,
    )
    print(f'Init model successful: {model}' )
    audit_elapsedtime(function="Initiating transcription model", start=start)
    return pipe

def init_model_ner():
    print("Initiating NER model...")
    start = set_start()
    model = GLiNER.from_pretrained("urchade/gliner_multi")
    audit_elapsedtime(function="Initiating NER model", start=start)
    return model

def transcribe (audio_sample: bytes, pipe) -> str:
    start = set_start()
    # dataset = load_dataset("distil-whisper/librispeech_long", "clean", split="validation")
    # sample = dataset[0]["audio"]
    result = pipe(audio_sample)
    audit_elapsedtime(function="Transcription", start=start)
    print(result)
    
    st.write('trancription: ', result["text"])
    return result["text"]

def get_entity_labels(model: GLiNER, text: str, labels: list): #-> Lead_labels:
    start = set_start()
    entities = model.predict_entities(text, labels)
    audit_elapsedtime(function="Retreiving entity labels from text", start=start)

    for entity in entities:
        print(entity["text"], "=>", entity["label"])
    st.write('Entities: ', entities)
    # return Lead_Labels()

if __name__ == "__main__":
    print("IN __name__")
    main()