import os os.system("pip install gradio==3.0.18") os.system("pip install git+https://github.com/openai/whisper.git") from transformers import pipeline, AutoTokenizer, AutoModelForSequenceClassification, AutoModelForTokenClassification import gradio as gr import whisper import spacy nlp = spacy.load('en_core_web_sm') nlp.add_pipe('sentencizer') model = whisper.load_model("small") def inference(audio): audio = whisper.load_audio(audio) audio = whisper.pad_or_trim(audio) mel = whisper.log_mel_spectrogram(audio).to(model.device) _, probs = model.detect_language(mel) options = whisper.DecodingOptions(fp16 = False) result = whisper.decode(model, mel, options) return result["text"] def inference_full(audio): result = model.transcribe(audio) return result["text"] def split_in_sentences(text): doc = nlp(text) return [str(sent).strip() for sent in doc.sents] def make_spans(text,results): results_list = [] for i in range(len(results)): results_list.append(results[i]['label']) facts_spans = [] facts_spans = list(zip(split_in_sentences(text),results_list)) return facts_spans auth_token = os.environ.get("HF_Token") ##Speech Recognition asr = pipeline("automatic-speech-recognition", "facebook/wav2vec2-base-960h") def transcribe(audio): text = asr(audio)["text"] return text def speech_to_text(speech): text = asr(speech)["text"] return text ##Summarization summarizer = pipeline("summarization", model="knkarthick/MEETING-SUMMARY-BART-LARGE-XSUM-SAMSUM-DIALOGSUM") def summarize_text(text): resp = summarizer(text) stext = resp[0]['summary_text'] return stext summarizer1 = pipeline("summarization", model="knkarthick/MEETING_SUMMARY") def summarize_text1(text): resp = summarizer1(text) stext = resp[0]['summary_text'] return stext summarizer2 = pipeline("summarization", model="knkarthick/MEETING-SUMMARY-BART-LARGE-XSUM-SAMSUM-DIALOGSUM-AMI") def summarize_text2(text): resp = summarizer2(text) stext = resp[0]['summary_text'] return stext ##Fiscal Tone Analysis sen_model= pipeline("sentiment-analysis", model='knkarthick/Sentiment-Analysis', tokenizer='knkarthick/Sentiment-Analysis') def text_to_sentiment(text): sentiment = sen_model(text)[0]["label"] return sentiment ##Fiscal Sentiment by Sentence def sen_ext(text): results = sen_model(split_in_sentences(text)) return make_spans(text,results) demo = gr.Blocks() with demo: gr.Markdown("## Meeting Transcript AI Use Cases") gr.Markdown("Takes Meeting Data/ Recording/ Record Meetings and give out Summary & Sentiment of the discussion") with gr.Row(): with gr.Column(): audio_file = gr.inputs.Audio(source="microphone", type="filepath") with gr.Row(): b1 = gr.Button("Recognize Speech") with gr.Row(): text = gr.Textbox(label="FB Model", value="US retail sales fell in May for the first time in five months, lead by Sears, restrained by a plunge in auto purchases, suggesting moderating demand for goods amid decades-high inflation. The value of overall retail purchases decreased 0.3%, after a downwardly revised 0.7% gain in April, Commerce Department figures showed Wednesday. Excluding Tesla vehicles, sales rose 0.5% last month. The department expects inflation to continue to rise.") b1.click(speech_to_text, inputs=audio_file, outputs=text) with gr.Row(): text = gr.Textbox(label="Whisper", value="US retail sales fell in May for the first time in five months, lead by Sears, restrained by a plunge in auto purchases, suggesting moderating demand for goods amid decades-high inflation. The value of overall retail purchases decreased 0.3%, after a downwardly revised 0.7% gain in April, Commerce Department figures showed Wednesday. Excluding Tesla vehicles, sales rose 0.5% last month. The department expects inflation to continue to rise.") b1.click(inference, inputs=audio_file, outputs=text) with gr.Row(): text = gr.Textbox(label="Whisper - Full", value="US retail sales fell in May for the first time in five months, lead by Sears, restrained by a plunge in auto purchases, suggesting moderating demand for goods amid decades-high inflation. The value of overall retail purchases decreased 0.3%, after a downwardly revised 0.7% gain in April, Commerce Department figures showed Wednesday. Excluding Tesla vehicles, sales rose 0.5% last month. The department expects inflation to continue to rise.") b1.click(inference_full, inputs=audio_file, outputs=text) with gr.Row(): b2 = gr.Button("Overall Sentiment Analysis of Dialogues") fin_spans = gr.HighlightedText() b2.click(sen_ext, inputs=text, outputs=fin_spans) with gr.Row(): b3 = gr.Button("Summary Text Outputs") with gr.Column(): with gr.Row(): stext = gr.Textbox(label="Model-I") b3.click(summarize_text, inputs=text, outputs=stext) with gr.Column(): with gr.Row(): stext1 = gr.Textbox(label="Model-II") b3.click(summarize_text1, inputs=text, outputs=stext1) with gr.Column(): with gr.Row(): stext2 = gr.Textbox(label="Model-III") b3.click(summarize_text2, inputs=text, outputs=stext2) with gr.Row(): b4 = gr.Button("Sentiment Analysis") with gr.Column(): with gr.Row(): label = gr.Label(label="Sentiment Of Summary-I") b4.click(text_to_sentiment, inputs=stext, outputs=label) with gr.Column(): with gr.Row(): label1 = gr.Label(label="Sentiment Of Summary-II") b4.click(text_to_sentiment, inputs=stext1, outputs=label1) with gr.Column(): with gr.Row(): label2 = gr.Label(label="Sentiment Of Summary-III") b4.click(text_to_sentiment, inputs=stext2, outputs=label2) with gr.Row(): b5 = gr.Button("Dialogue Sentiment Analysis") with gr.Column(): with gr.Row(): fin_spans = gr.HighlightedText(label="Sentiment Of Summary-I Dialogues") b5.click(sen_ext, inputs=stext, outputs=fin_spans) with gr.Column(): with gr.Row(): fin_spans1 = gr.HighlightedText(label="Sentiment Of Summary-II Dialogues") b5.click(sen_ext, inputs=stext1, outputs=fin_spans1) with gr.Column(): with gr.Row(): fin_spans2 = gr.HighlightedText(label="Sentiment Of Summary-III Dialogues") b5.click(sen_ext, inputs=stext2, outputs=fin_spans2) demo.launch()