app.py

import os
import re
import functools
from functools import partial

import requests
import pandas as pd
import plotly.express as px

import torch
import gradio as gr
from transformers import pipeline, Wav2Vec2ProcessorWithLM
from pyannote.audio import Pipeline
import whisperx

from utils import split_into_sentences, sentiment, color_map
from utils import speech_to_text as stt

os.environ["TOKENIZERS_PARALLELISM"] = "false"
device = 0 if torch.cuda.is_available() else -1


# Audio components
whisper_device = "cuda" if torch.cuda.is_available() else "cpu"
whisper = whisperx.load_model("tiny.en", whisper_device)
alignment_model, metadata = whisperx.load_align_model(language_code="en", device=whisper_device)
speaker_segmentation = Pipeline.from_pretrained("pyannote/[email protected]",
                                    use_auth_token=os.environ['ENO_TOKEN'])


# Text components
emotion_pipeline = pipeline(
    "text-classification",
    model="bhadresh-savani/distilbert-base-uncased-emotion",
    device=device,
)

EXAMPLES = [["Customer_Support_Call.wav"]]


speech_to_text = partial(
    stt, 
    speaker_segmentation=speaker_segmentation, 
    whisper=whisper, 
    alignment_model=alignment_model, 
    metadata=metadata, 
    whisper_device=whisper_device
    )

with gr.Blocks() as demo:

    with gr.Row():
        with gr.Column():
            audio = gr.Audio(label="Audio file", type="filepath")
            btn = gr.Button("Transcribe and Diarize")

            gr.Markdown("**Call Transcript:**")
            diarized = gr.HighlightedText(label="Call Transcript")
            sentiment_btn = gr.Button("Get Customer Sentiment")
            analyzed = gr.HighlightedText(color_map=color_map)
            plot = gr.Plot(label="Sentiment over time", type="plotly")

        with gr.Column():
            gr.Markdown("## Example Files")
            gr.Examples(
                examples=EXAMPLES,
                inputs=[audio],
                outputs=[diarized],
                fn=speech_to_text,
                cache_examples=True
            )
    # when example button is clicked, convert audio file to text and diarize
    btn.click(
        fn=speech_to_text,
        inputs=audio,
        outputs=diarized,
    )

    # when sentiment button clicked, display highlighted text and plot
    sentiment_btn.click(fn=partial(sentiment, emotion_pipeline=emotion_pipeline), inputs=diarized, outputs=[analyzed, plot])


demo.launch(debug=1)