Spaces:
Build error
Build error
File size: 6,156 Bytes
bb4dfd1 20ec090 bb4dfd1 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 |
import re
import functools
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
from librosa import load, resample
import whisperx
import re
alphabets= "([A-Za-z])"
prefixes = "(Mr|St|Mrs|Ms|Dr)[.]"
suffixes = "(Inc|Ltd|Jr|Sr|Co)"
starters = "(Mr|Mrs|Ms|Dr|He\s|She\s|It\s|They\s|Their\s|Our\s|We\s|But\s|However\s|That\s|This\s|Wherever)"
acronyms = "([A-Z][.][A-Z][.](?:[A-Z][.])?)"
websites = "[.](com|net|org|io|gov)"
def split_into_sentences(text):
text = " " + text + " "
text = text.replace("\n"," ")
text = re.sub(prefixes,"\\1<prd>",text)
text = re.sub(websites,"<prd>\\1",text)
if "Ph.D" in text: text = text.replace("Ph.D.","Ph<prd>D<prd>")
text = re.sub("\s" + alphabets + "[.] "," \\1<prd> ",text)
text = re.sub(acronyms+" "+starters,"\\1<stop> \\2",text)
text = re.sub(alphabets + "[.]" + alphabets + "[.]" + alphabets + "[.]","\\1<prd>\\2<prd>\\3<prd>",text)
text = re.sub(alphabets + "[.]" + alphabets + "[.]","\\1<prd>\\2<prd>",text)
text = re.sub(" "+suffixes+"[.] "+starters," \\1<stop> \\2",text)
text = re.sub(" "+suffixes+"[.]"," \\1<prd>",text)
text = re.sub(" " + alphabets + "[.]"," \\1<prd>",text)
if "”" in text: text = text.replace(".”","”.")
if "\"" in text: text = text.replace(".\"","\".")
if "!" in text: text = text.replace("!\"","\"!")
if "?" in text: text = text.replace("?\"","\"?")
text = text.replace(".",".<stop>")
text = text.replace("?","?<stop>")
text = text.replace("!","!<stop>")
text = text.replace("<prd>",".")
sentences = text.split("<stop>")
sentences = sentences[:-1]
sentences = [s.strip() for s in sentences]
return sentences
def summarize(diarized, check, summarization_pipeline):
"""
diarized: a list of tuples. Each tuple has a string to be displayed and a label for highlighting.
The start/end times are not highlighted [(speaker text, speaker id), (start time/end time, None)]
check is a list of speaker ids whose speech will get summarized
"""
if not check:
return ""
# Combine text based on the speaker id
text_lines = [f"{d[1]}: {d[0]}" if len(check) == 2 and d[1] is not None else d[0] for d in diarized if d[1] in check]
text = "\n".join(text_lines)
# Cache the inner function because the outer function cannot be cached
@functools.lru_cache(maxsize=128)
def call_summarize_api(text):
return summarization_pipeline(text)[0]["summary_text"]
return call_summarize_api(text)
# display if the sentiment value is above these thresholds
thresholds = {
"joy": 0.99,
"anger": 0.95,
"surprise": 0.95,
"sadness": 0.98,
"fear": 0.95,
"love": 0.99,
}
color_map = {
"joy": "green",
"anger": "red",
"surprise": "yellow",
"sadness": "blue",
"fear": "orange",
"love": "purple",
}
def sentiment(diarized, emotion_pipeline):
def split_into_intervals(speaker_speech, start_time, end_time):
sentences = split_into_sentences(speaker_speech)
interval_size = (end_time - start_time) / len(sentences)
return sentences, interval_size
def process_customer_emotion(outputs, sentences, start_time, interval_size):
sentiments = []
for idx, (o, t) in enumerate(zip(outputs, sentences)):
sent = "neutral"
if o["score"] > thresholds[o["label"]]:
sentiments.append((t + f"({round(idx*interval_size+start_time,1)} s)", o["label"]))
if o["label"] in {"joy", "love", "surprise"}:
sent = "positive"
elif o["label"] in {"sadness", "anger", "fear"}:
sent = "negative"
if sent != "neutral":
to_plot.append((start_time + idx * interval_size, sent))
plot_sentences.append(t)
return sentiments
x_min = 100
x_max = 0
customer_sentiments, to_plot, plot_sentences = [], [], []
for i in range(0, len(diarized), 2):
speaker_speech, speaker_id = diarized[i]
times, _ = diarized[i + 1]
start_time, end_time = map(float, times[5:].split("-"))
x_min, x_max = min(x_min, start_time), max(x_max, end_time)
if "Customer" in speaker_id:
sentences, interval_size = split_into_intervals(speaker_speech, start_time, end_time)
outputs = emotion_pipeline(sentences)
customer_sentiments.extend(process_customer_emotion(outputs, sentences, start_time, interval_size))
plot_df = pd.DataFrame(data={"x": [x for x, _ in to_plot], "y": [y for _, y in to_plot], "sentence": plot_sentences})
fig = px.line(plot_df, x="x", y="y", hover_data={"sentence": True, "x": True, "y": False}, labels={"x": "time (seconds)", "y": "sentiment"}, title=f"Customer sentiment over time", markers=True)
fig.update_yaxes(categoryorder="category ascending")
fig.update_layout(font=dict(size=18), xaxis_range=[x_min - 5, x_max + 5])
return customer_sentiments, fig
def speech_to_text(speech_file, speaker_segmentation, whisper, alignment_model, metadata, whisper_device):
def process_chunks(turn, chunks):
diarized = ""
i = 0
while i < len(chunks) and chunks[i]["end"] <= turn.end:
diarized += chunks[i]["text"] + " "
i += 1
return diarized, i
speaker_output = speaker_segmentation(speech_file)
result = whisper.transcribe(speech_file)
chunks = whisperx.align(result["segments"], alignment_model, metadata, speech_file, whisper_device)["word_segments"]
diarized_output = []
i = 0
speaker_counter = 0
for turn, _, _ in speaker_output.itertracks(yield_label=True):
speaker = "Customer" if speaker_counter % 2 == 0 else "Support"
diarized, i = process_chunks(turn, chunks[i:])
if diarized:
diarized_output.extend([(diarized, speaker), (f"from {turn.start:.2f}-{turn.end:.2f}", None)])
speaker_counter += 1
return diarized_output |