streamlite / app.py
Sajjo's picture
Update app.py
6d4ade5 verified
import os
import re
import warnings
import gradio as gr
from transformers import pipeline, AutoProcessor
from pyctcdecode import build_ctcdecoder
from transformers import Wav2Vec2ProcessorWithLM
from indictrans import Transliterator
# Initialize ASR pipelines
asr_models = {
"Odiya": pipeline(task="automatic-speech-recognition", model="cdactvm/w2v-bert-2.0-odia_v1"),
"Odiya-trans": pipeline(task="automatic-speech-recognition", model="cdactvm/w2v-bert-2.0-odia_v1"),
"Hindi": pipeline(task="automatic-speech-recognition", model="cdactvm/w2v-bert-2.0-hindi_v1"),
"Hindi-trans": pipeline(task="automatic-speech-recognition", model="cdactvm/w2v-bert-2.0-hindi_v1"),
# Add other models similarly
# "Kannada": pipeline(...),
# "Telugu": pipeline(...),
# "Bangala": pipeline(...),
"Assamese-Model2": pipeline(task="automatic-speech-recognition", model="cdactvm/w2v-assames"),
}
# Initialize Assamese model with Language Model
processor = AutoProcessor.from_pretrained("cdactvm/w2v-assames")
vocab_dict = processor.tokenizer.get_vocab()
sorted_vocab_dict = {k.lower(): v for k, v in sorted(vocab_dict.items(), key=lambda item: item[1])}
decoder = build_ctcdecoder(labels=list(sorted_vocab_dict.keys()), kenlm_model_path="lm.binary")
processor_with_lm = Wav2Vec2ProcessorWithLM(feature_extractor=processor.feature_extractor,
tokenizer=processor.tokenizer,
decoder=decoder)
asr_models["Assamese-LM"] = pipeline("automatic-speech-recognition", model="cdactvm/w2v-assames",
tokenizer=processor_with_lm,
feature_extractor=processor_with_lm.feature_extractor,
decoder=processor_with_lm.decoder)
# Initialize Transliterator
transliterators = {
"Odiya-trans": Transliterator(source='ori', target='eng', build_lookup=True),
"Hindi-trans": Transliterator(source='hin', target='eng', build_lookup=True),
# Add other transliterators similarly
}
# Function to clean HTML tags from text
def cleanhtml(raw_html):
return re.sub(r'<.*?>', '', raw_html)
# Transcribe audio using the selected model
def transcribe(lng, speech, transliterate=False):
model = asr_models.get(lng)
if not model:
return f"Unsupported language: {lng}"
result = model(speech)
text = result.get("text")
if text is None:
return "Error: ASR returned None"
if transliterate:
trn = transliterators.get(lng + "-trans")
if not trn:
return f"Transliterator not available for: {lng}"
sentence = trn.transform(text)
if sentence is None:
return "Error: Transliteration returned None"
return process_transcription(sentence)
return cleanhtml(text)
# Function to process and correct transcriptions
def process_transcription(sentence):
replaced_words = replace_words(sentence)
processed_sentence = process_doubles(replaced_words)
return convert_to_text(processed_sentence)
# Replace incorrectly spelled words
def replace_words(sentence):
replacements = [
(r'\bjiro\b', 'zero'), (r'\bjero\b', 'zero'),
(r'\bnn\b', 'one'), (r'\bn\b', 'one'), (r'\bvan\b', 'one'), (r'\bna\b', 'one'), (r'\bek\b', 'one'),
(r'\btu\b', 'two'), (r'\btoo\b', 'two'), (r'\bdo\b', 'two'),
(r'\bthiri\b', 'three'), (r'\btiri\b', 'three'), (r'\bdubalathri\b', 'double three'), (r'\btin\b', 'three'),
(r'\bfor\b', 'four'), (r'\bfore\b', 'four'),
(r'\bfib\b', 'five'), (r'\bpaanch\b', 'five'),
(r'\bchha\b', 'six'), (r'\bchhah\b', 'six'), (r'\bchau\b', 'six'),
(r'\bdublseven\b', 'double seven'), (r'\bsath\b', 'seven'),
(r'\baath\b', 'eight'),
(r'\bnau\b', 'nine'),
(r'\bdas\b', 'ten'),
(r'\bnineeit\b', 'nine eight'),
(r'\bfipeit\b', 'five eight'), (r'\bdubal\b', 'double'), (r'\bsevenatu\b', 'seven two'),
]
for pattern, replacement in replacements:
sentence = re.sub(pattern, replacement, sentence)
return sentence
# Process "double" followed by a number
def process_doubles(sentence):
tokens = sentence.split()
result = []
i = 0
while i < len(tokens):
if tokens[i] in ("double", "dubal") and i + 1 < len(tokens):
result.extend([tokens[i + 1]] * 2)
i += 2
else:
result.append(tokens[i])
i += 1
return ' '.join(result)
# Convert Soundex code back to text
def convert_to_text(input_sentence):
word_to_code_map = {}
transcript = sentence_to_transcript(input_sentence, word_to_code_map)
if transcript is None:
return "Error: Transcript conversion returned None"
numbers = text2int(transcript)
if numbers is None:
return "Error: Text to number conversion returned None"
code_to_word_map = {v: k for k, v in word_to_code_map.items()}
return transcript_to_sentence(numbers, code_to_word_map)
# Convert text to numerical representation
def text2int(textnum, numwords={}):
units = ['Z600', 'O500', 'T000', 'T600', 'F600', 'F100', 'S220', 'S150', 'E300', 'N500',
'T500', 'E415', 'T410', 'T635', 'F635', 'F135', 'S235', 'S153', 'E235', 'N535']
tens = ['', '', 'T537', 'T637', 'F637', 'F137', 'S230', 'S153', 'E230', 'N530']
scales = ['H536', 'T253', 'M450', 'C600']
ordinal_words = {'oh': 'Z600', 'first': 'O500', 'second': 'T000', 'third': 'T600', 'fourth': 'F600', 'fifth': 'F100',
'sixth': 'S200', 'seventh': 'S150', 'eighth': 'E230', 'ninth': 'N500', 'twelfth': 'T410'}
ordinal_endings = [('ieth', 'y'), ('th', '')]
if not numwords:
numwords['and'] = (1, 0)
for idx, word in enumerate(units): numwords[word] = (1, idx)
for idx, word in enumerate(tens): numwords[word] = (1, idx * 10)
for idx, word in enumerate(scales): numwords[word] = (10 ** (idx * 3 or 2), 0)
textnum = textnum.replace('-', ' ')
current = result = 0
curstring = ''
onnumber = False
lastunit = False
lastscale = False
def is_numword(x):
if is_number(x):
return True
if word in numwords:
return True
return False
def from_numword(x):
if is_number(x):
scale = 0
increment = int(x.replace(',', ''))
return scale, increment
return numwords[x]
for word in textnum.split():
if word in ordinal_words:
scale, increment = (1, ordinal_words[word])
current = current * scale + increment
if scale > 100:
result += current
current = 0
lastunit = True
lastscale = False
onnumber = True
else:
for ending, replacement in ordinal_endings:
if word.endswith(ending):
word = "%s%s" % (word[:-len(ending)], replacement)
if not is_numword(word) or (word == 'and' and not lastscale):
if onnumber:
curstring += repr(result + current) + " "
curstring += word
if word[-1] != '-':
curstring += " "
result = current = 0
onnumber = False
lastunit = False
lastscale = False
else:
scale, increment = from_numword(word)
onnumber = True
if lastunit and (word in units or word in ordinal_words):
curstring += repr(result + current)
result = current = 0
if scale > 1:
current = max(1, current)
current = current * scale + increment
if scale > 100:
result += current
current = 0
lastunit = word in units
lastscale = word in scales
if onnumber:
curstring += repr(result + current)
return curstring
# Check if a word is a number
def is_number(s):
try:
float(s.replace(',', ''))
return True
except ValueError:
return False
# Convert sentence to transcript using Soundex
def sentence_to_transcript(sentence, word_to_code_map):
with warnings.catch_warnings():
warnings.simplefilter("ignore")
from metaphone import doublemetaphone
transcript = []
for word in sentence.split():
codes = doublemetaphone(word)
word_code = next((code for code in codes if code), None)
if not word_code:
continue
if word_code not in word_to_code_map:
word_to_code_map[word] = word_code
transcript.append(word_code)
return ' '.join(transcript)
# Convert transcript back to sentence using Soundex
def transcript_to_sentence(transcript, code_to_word_map):
sentence = []
for code in transcript.split():
word = code_to_word_map.get(code, '')
if word:
sentence.append(word)
return ' '.join(sentence)
# Set theme colors for Gradio interface
theme_colors = {
"bg_color": "#0E1117",
"bg_secondary_color": "#161A25",
"input_color": "#161A25",
"input_text_color": "#C0C0BF",
"button_color": "#4A6AF2",
"button_primary_text_color": "#FFFFFF",
"button_secondary_color": "#A0A0A0",
"button_secondary_text_color": "#000000"
}
# Apply theme to Gradio blocks
def apply_theme(demo):
demo.set_theme({
"background_color": theme_colors["bg_color"],
"secondary_background_color": theme_colors["bg_secondary_color"],
"input_background_color": theme_colors["input_color"],
"input_text_color": theme_colors["input_text_color"],
"button_primary_background_color": theme_colors["button_color"],
"button_primary_text_color": theme_colors["button_primary_text_color"],
"button_secondary_background_color": theme_colors["button_secondary_color"],
"button_secondary_text_color": theme_colors["button_secondary_text_color"]
})
# Create Gradio interface
with gr.Blocks() as demo:
apply_theme(demo)
gr.Markdown("<h1><center>Test</center></h1>")
with gr.Row():
language = gr.Dropdown(list(asr_models.keys()), label="Language", value="Hindi")
speech_input = gr.Audio(source="microphone", type="filepath", label="Speech")
text_output = gr.Textbox(label="Output")
submit_btn = gr.Button("Submit")
def process_audio(lang, speech):
transliterate = lang.endswith("-trans")
return transcribe(lang, speech, transliterate)
submit_btn.click(process_audio, inputs=[language, speech_input], outputs=text_output)
# Launch the Gradio app on a different port
demo.launch(server_port=7861)