app.py

import gradio as gr
import torch
import librosa
import time
import pandas as pd
from datetime import datetime
from transformers import Wav2Vec2ForSequenceClassification, Wav2Vec2FeatureExtractor

DESCRIPTION = "Store a record of previous calls in order to verify if the client already called or not. Pretrained on `https://huggingface.co/datasets/superb` using [S3PRL recipe](https://github.com/s3prl/s3prl/tree/master/s3prl/downstream/voxceleb1)."

# COLUMNS = ["call_id", "date", "client_id", "duration", "new"]
model = Wav2Vec2ForSequenceClassification.from_pretrained("superb/wav2vec2-large-superb-sid")
feature_extractor = Wav2Vec2FeatureExtractor.from_pretrained("superb/wav2vec2-large-superb-sid")

def file_to_array(path):
    speech, _ = librosa.load(path, sr=16000, mono=True)
    duration = librosa.get_duration(y=speech)
    return speech, duration


def handler(audio_path):
    calls = pd.read_csv("call_records.csv")
    speech, duration = file_to_array(audio_path)

    # compute attention masks and normalize the waveform if needed
    inputs = feature_extractor(speech, sampling_rate=16000, padding=True, return_tensors="pt")

    logits = model(**inputs).logits
    predicted_ids = torch.argmax(logits, dim=-1)
    labels = [model.config.id2label[_id] for _id in predicted_ids.tolist()]

    client_id = labels[0]    
    call_id = str(int(time.time()))

    date = datetime.now().strftime("%d/%m/%Y %H:%M:%S")

    n_of_calls = len(calls.loc[calls.client_id == client_id])
    new = n_of_calls == 0

    # add new call record 
    record = [call_id, date, client_id, duration, new]
    calls.loc[len(calls)] = record

    calls.to_csv("call_records.csv", index=False)

    if new:
        return f"New client call: Client ID {client_id}"
    
    return f"Client {client_id} calling again: {n_of_calls} previous calls"
    

first = gr.Interface(
    fn=handler,
    inputs=gr.Audio(label="Speech Audio", type="filepath"),
       
    outputs=gr.Text(label="Output", value="..."),
    description=DESCRIPTION

)

second = gr.Interface(
    fn=handler,
    inputs=gr.Audio(label="Microphone Input", source="microphone", type="filepath"),
    outputs=gr.Text(label="Output", value="..."),
    description=DESCRIPTION
)

app = gr.TabbedInterface(
    [first, second],
    title="Speaker Call Verification 🎤",
    tab_names=["Audio Upload", "Microphone"],
)
app.launch()