# import warnings # warnings.filterwarnings("ignore") # import os # import re # import gradio as gr # import numpy as np # import torchaudio # import nbimporter # from transformers import pipeline # from transformers import AutoProcessor # from pyctcdecode import build_ctcdecoder # from transformers import Wav2Vec2ProcessorWithLM # from text2int import text_to_int # from isNumber import is_number # from Text2List import text_to_list # from convert2list import convert_to_list # from processDoubles import process_doubles # from replaceWords import replace_words # # transcriber = pipeline(task="automatic-speech-recognition", model="cdactvm/w2v-bert-2.0-hindi_v1") # # processor = AutoProcessor.from_pretrained("cdactvm/w2v-bert-2.0-hindi_v1") # # 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 # # ) # # processor.feature_extractor._processor_class = "Wav2Vec2ProcessorWithLM" # def transcribe(audio): # # # Process the audio file # transcript = transcriber(audio) # text_value = transcript['text'] # print(text_value) # processd_doubles=process_doubles(text_value) # converted_to_list=convert_to_list(processd_doubles,text_to_list()) # replaced_words = replace_words(converted_to_list) # converted_text=text_to_int(replaced_words) # return converted_text # # demo = gr.Interface( # # transcribe, # # gr.Audio(sources="microphone", type="filepath"), # # "text", # # ) # # demo.launch() # demo=gr.Interface( # transcribe, # inputs=[ # gr.Audio(sources=["microphone","upload"], type="filepath"), # ], # outputs=[ # "textbox" # ], # title="Automatic Speech Recognition", # description = "Demo for Automatic Speech Recognition. Use microphone to record speech. Please press Record button. Initially it will take some time to load the model. The recognized text will appear in the output textbox", # ).launch() ############################################################### import warnings warnings.filterwarnings("ignore") import os import re import gradio as gr import numpy as np import torchaudio import nbimporter from transformers import pipeline from transformers import AutoProcessor from pyctcdecode import build_ctcdecoder from transformers import Wav2Vec2ProcessorWithLM from text2int import text_to_int from isNumber import is_number from Text2List import text_to_list from convert2list import convert_to_list from processDoubles import process_doubles from replaceWords import replace_words hindi_model = pipeline(task="automatic-speech-recognition", model="cdactvm/w2v-bert-2.0-hindi_v1") import warnings import gradio as gr from transformers import pipeline from transformers import AutoProcessor from pyctcdecode import build_ctcdecoder from transformers import Wav2Vec2ProcessorWithLM import os import re #import torchaudio # Initialize the speech recognition pipeline and transliterator odia_model1 = pipeline(task="automatic-speech-recognition", model="cdactvm/w2v-bert-odia_v1") odia_model2 = pipeline(task="automatic-speech-recognition", model="cdactvm/w2v-bert-odia_v2") # p2 = pipeline(task="automatic-speech-recognition", model="cdactvm/w2v-bert-2.0-hindi_v1") # punjaib_modle_30000=pipeline(task="automatic-speech-recognition", model="cdactvm/wav2vec-bert-punjabi-30000-model") # punjaib_modle_155750=pipeline(task="automatic-speech-recognition", model="cdactvm/wav2vec-bert-punjabi-155750-model") # punjaib_modle_70000_aug=pipeline(task="automatic-speech-recognition", model="cdactvm/w2v-bert-model-30000-augmented") #p3 = pipeline(task="automatic-speech-recognition", model="cdactvm/kannada_w2v-bert_model") #p4 = pipeline(task="automatic-speech-recognition", model="cdactvm/telugu_w2v-bert_model") #p5 = pipeline(task="automatic-speech-recognition", model="Sajjo/w2v-bert-2.0-bangala-gpu-CV16.0_v2") #p6 = pipeline(task="automatic-speech-recognition", model="cdactvm/hf-open-assames") # p7 = pipeline(task="automatic-speech-recognition", model="cdactvm/w2v-assames") processor = AutoProcessor.from_pretrained("cdactvm/w2v-bert-odia_v2") 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 ) processor.feature_extractor._processor_class = "Wav2Vec2ProcessorWithLM" #p8 = pipeline("automatic-speech-recognition", model="cdactvm/w2v-assames", tokenizer=processor_with_lm, feature_extractor=processor_with_lm.feature_extractor, decoder=processor_with_lm.decoder) os.system('git clone https://github.com/irshadbhat/indic-trans.git') os.system('pip install ./indic-trans/.') #HF_TOKEN = os.getenv('HF_TOKEN') #hf_writer = gr.HuggingFaceDatasetSaver(HF_TOKEN, "asr_demo") from indictrans import Transliterator ########################################### # Function to 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'\bnn\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 # Function to 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"): if i + 1 < len(tokens): result.append(tokens[i + 1]) result.append(tokens[i + 1]) i += 2 else: result.append(tokens[i]) i += 1 else: result.append(tokens[i]) i += 1 return ' '.join(result) # Function to generate Soundex code for a word def soundex(word): word = word.upper() word = ''.join(filter(str.isalpha, word)) if not word: return None soundex_mapping = { 'B': '1', 'F': '1', 'P': '1', 'V': '1', 'C': '2', 'G': '2', 'J': '2', 'K': '2', 'Q': '2', 'S': '2', 'X': '2', 'Z': '2', 'D': '3', 'T': '3', 'L': '4', 'M': '5', 'N': '5', 'R': '6' } soundex_code = word[0] for char in word[1:]: if char not in ('H', 'W'): soundex_code += soundex_mapping.get(char, '0') soundex_code = soundex_code[0] + ''.join(c for i, c in enumerate(soundex_code[1:]) if c != soundex_code[i]) soundex_code = soundex_code.replace('0', '') + '000' return soundex_code[:4] # Function to convert text to numerical representation def is_number(x): if type(x) == str: x = x.replace(',', '') try: float(x) except: return False return True 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 x 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 onnumber = True lastunit = False lastscale = False 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 + " " result = current = 0 onnumber = False lastunit = False lastscale = False else: scale, increment = from_numword(word) onnumber = True if lastunit and (word not in scales): 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 lastscale = False lastunit = False if word in scales: lastscale = True elif word in units: lastunit = True if onnumber: curstring += repr(result + current) return curstring # Convert sentence to transcript using Soundex def sentence_to_transcript(sentence, word_to_code_map): words = sentence.split() transcript_codes = [] for word in words: if word not in word_to_code_map: word_to_code_map[word] = soundex(word) transcript_codes.append(word_to_code_map[word]) transcript = ' '.join(transcript_codes) return transcript # Convert transcript back to sentence using mapping def transcript_to_sentence(transcript, code_to_word_map): codes = transcript.split() sentence_words = [] for code in codes: sentence_words.append(code_to_word_map.get(code, code)) sentence = ' '.join(sentence_words) return sentence # # Process the audio file # transcript = pipe("./odia_recorded/AUD-20240614-WA0004.wav") # text_value = transcript['text'] # sentence = trn.transform(text_value) # replaced_words = replace_words(sentence) # processed_sentence = process_doubles(replaced_words) # input_sentence_1 = processed_sentence # Create empty mappings word_to_code_map = {} code_to_word_map = {} # Convert sentence to transcript # transcript_1 = sentence_to_transcript(input_sentence_1, word_to_code_map) # Convert transcript to numerical representation # numbers = text2int(transcript_1) # Create reverse mapping code_to_word_map = {v: k for k, v in word_to_code_map.items()} def process_transcription(input_sentence): word_to_code_map = {} code_to_word_map = {} transcript_1 = sentence_to_transcript(input_sentence, word_to_code_map) if transcript_1 is None: return "Error: Transcript conversion returned None" numbers = text2int(transcript_1) 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()} text = transcript_to_sentence(numbers, code_to_word_map) return text ########################################### def transcribe_punjabi_30000(speech): text = punjaib_modle_30000(speech)["text"] text = text.replace("[PAD]","") if text is None: return "Error: ASR returned None" return text def transcribe_punjabi_eng_model_30000(speech): trn = Transliterator(source='pan', target='eng', build_lookup=True) text = punjaib_modle_30000(speech)["text"] text = text.replace("[PAD]","") if text is None: return "Error: ASR returned None" sentence = trn.transform(text) if sentence is None: return "Error: Transliteration returned None" replaced_words = replace_words(sentence) processed_sentence = process_doubles(replaced_words) return process_transcription(processed_sentence) return sentence def transcribe_punjabi_70000_aug(speech): text = punjaib_modle_70000_aug(speech)["text"] text = text.replace("","") if text is None: return "Error: ASR returned None" return text def transcribe_punjabi_eng_model_70000_aug(speech): trn = Transliterator(source='pan', target='eng', build_lookup=True) text = punjaib_modle_70000_aug(speech)["text"] text = text.replace("","") if text is None: return "Error: ASR returned None" sentence = trn.transform(text) if sentence is None: return "Error: Transliteration returned None" replaced_words = replace_words(sentence) processed_sentence = process_doubles(replaced_words) return process_transcription(processed_sentence) return sentence def transcribe_punjabi_155750(speech): text = punjaib_modle_155750(speech)["text"] text = text.replace("[PAD]","") if text is None: return "Error: ASR returned None" return text def transcribe_punjabi_eng_model_155750(speech): trn = Transliterator(source='pan', target='eng', build_lookup=True) text = punjaib_modle_155750(speech)["text"] text = text.replace("[PAD]","") if text is None: return "Error: ASR returned None" sentence = trn.transform(text) if sentence is None: return "Error: Transliteration returned None" replaced_words = replace_words(sentence) processed_sentence = process_doubles(replaced_words) return process_transcription(processed_sentence) return sentence ########################################### def transcribe_odiya_model1(speech): text = odia_model1(speech)["text"] if text is None: return "Error: ASR returned None" return text def transcribe_odiya_model2(speech): text = odia_model2(speech)["text"] if text is None: return "Error: ASR returned None" return text def transcribe_odiya_eng_model1(speech): trn = Transliterator(source='ori', target='eng', build_lookup=True) text = odia_model1(speech)["text"] if text is None: return "Error: ASR returned None" sentence = trn.transform(text) if sentence is None: return "Error: Transliteration returned None" replaced_words = replace_words(sentence) processed_sentence = process_doubles(replaced_words) return process_transcription(processed_sentence) def transcribe_odiya_eng_model2(speech): trn = Transliterator(source='ori', target='eng', build_lookup=True) text = odia_model2(speech)["text"] if text is None: return "Error: ASR returned None" sentence = trn.transform(text) if sentence is None: return "Error: Transliteration returned None" replaced_words = replace_words(sentence) processed_sentence = process_doubles(replaced_words) return process_transcription(processed_sentence) ######################################## def cleanhtml(raw_html): cleantext = re.sub(r'<.*?>', '', raw_html) return cleantext ####################################### # def transcribe_hindi(speech): # text = p2(speech)["text"] # if text is None: # return "Error: ASR returned None" # return text def transcribe_hindi(speech): text = hindi_model(speech)["text"] if text is None: return "Error: ASR returned None" hindi_map = { "सेवन": "7", "जीरो": "0", "वन" : "1", "टू" : "2", "थ्री" : "3", "त्री" : "3", "फोर" : "4", "फाइव": "5", "सिक्स": "6", "एट": "8", "नाइन": "9", "टेन": "10", "एक": "1", "दो": "2", "तीन": "3", "चार": "4", "पांच": "5", "पाँच": "5", "छह": "6", "छः": "6", "सात": "7", "आठ": "8", "नौ": "9", "दस": "10" } for hindi, num in hindi_map.items(): text = text.replace(hindi, num) # Split the string into parts separated by spaces parts = text.split(' ') # Initialize an empty list to store the processed parts processed_parts = [] # Iterate over each part for part in parts: # Check if the part is a number (contains only digits) if part.isdigit(): # If the previous part was also a number, concatenate them if processed_parts and processed_parts[-1].isdigit(): processed_parts[-1] += part else: processed_parts.append(part) else: # If the part is not a number, add it to the list as is processed_parts.append(part) # Join the processed parts back into a string with spaces text = ' '.join(processed_parts) return text ########################################################### def transcribe_kannada(speech): text = p3(speech)["text"] if text is None: return "Error: ASR returned None" return text def transcribe_telugu(speech): text = p4(speech)["text"] if text is None: return "Error: ASR returned None" return text def transcribe_bangala(speech): text = p5(speech)["text"] if text is None: return "Error: ASR returned None" return text def transcribe_assamese_LM(speech): text = p8(speech)["text"] text = cleanhtml(text) if text is None: return "Error: ASR returned None" return text def transcribe_assamese_model2(speech): text = p7(speech)["text"] text = cleanhtml(text) if text is None: return "Error: ASR returned None" return text def transcribe_ban_eng(speech): trn = Transliterator(source='ben', target='eng', build_lookup=True) text = p5(speech)["text"] if text is None: return "Error: ASR returned None" sentence = trn.transform(text) if sentence is None: return "Error: Transliteration returned None" replaced_words = replace_words(sentence) processed_sentence = process_doubles(replaced_words) return process_transcription(processed_sentence) def transcribe_hin_eng(speech): trn = Transliterator(source='hin', target='eng', build_lookup=True) text = p2(speech)["text"] if text is None: return "Error: ASR returned None" sentence = trn.transform(text) if sentence is None: return "Error: Transliteration returned None" replaced_words = replace_words(sentence) processed_sentence = process_doubles(replaced_words) return process_transcription(processed_sentence) def transcribe_kan_eng(speech): trn = Transliterator(source='kan', target='eng', build_lookup=True) text = p3(speech)["text"] if text is None: return "Error: ASR returned None" sentence = trn.transform(text) if sentence is None: return "Error: Transliteration returned None" replaced_words = replace_words(sentence) processed_sentence = process_doubles(replaced_words) return process_transcription(processed_sentence) def transcribe_tel_eng(speech): trn = Transliterator(source='tel', target='eng', build_lookup=True) text = p4(speech)["text"] if text is None: return "Error: ASR returned None" sentence = trn.transform(text) if sentence is None: return "Error: Transliteration returned None" replaced_words = replace_words(sentence) processed_sentence = process_doubles(replaced_words) return process_transcription(processed_sentence) def sel_lng(lng, mic=None, file=None): if mic is not None: audio = mic elif file is not None: audio = file else: return "You must either provide a mic recording or a file" if lng == "Odiya": return transcribe_odiya(audio) elif lng == "Odiya-trans": return transcribe_odiya_eng(audio) elif lng == "Hindi-trans": return transcribe_hin_eng(audio) elif lng == "Hindi": return transcribe_hindi(audio) elif lng == "Kannada-trans": return transcribe_kan_eng(audio) elif lng == "Kannada": return transcribe_kannada(audio) elif lng == "Telugu-trans": return transcribe_tel_eng(audio) elif lng == "Telugu": return transcribe_telugu(audio) elif lng == "Bangala-trans": return transcribe_ban_eng(audio) elif lng == "Bangala": return transcribe_bangala(audio) elif lng == "Assamese-LM": return transcribe_assamese_LM(audio) elif lng == "Assamese-Model2": return transcribe_assamese_model2(audio) elif lng == "Odia_model1": return transcribe_odiya_model1(audio) elif lng == "Odiya_trans_model1": return transcribe_odiya_eng_model1(audio) elif lng == "Odia_model2": return transcribe_odiya_model2(audio) elif lng == "Odia_trans_model2": return transcribe_odiya_eng_model2(audio) elif lng == "Punjabi_Model0": return transcribe_punjabi_30000(audio) elif lng == "Punjabi_Model0_Trans": return transcribe_punjabi_eng_model_30000(audio) elif lng == "Punjabi_Model_aug": return transcribe_punjabi_70000_aug(audio) elif lng == "Punjabi_Model_aug_Trans": return transcribe_punjabi_eng_model_70000_aug(audio) elif lng == "Punjabi_Model1": return transcribe_punjabi_155750(audio) elif lng == "Punjabi_Model1_Trans": return transcribe_punjabi_eng_model_155750(audio) # Convert transcript back to sentence # reconstructed_sentence_1 = transcript_to_sentence(numbers, code_to_word_map) # demo=gr.Interface( # fn=sel_lng, # inputs=[ # gr.Dropdown(["Hindi","Hindi-trans","Odiya","Odiya-trans"],value="Hindi",label="Select Language"), # gr.Audio(source="microphone", type="filepath"), # gr.Audio(source= "upload", type="filepath"), # #gr.Audio(sources="upload", type="filepath"), # #"state" # ], # outputs=[ # "textbox" # # #"state" # ], # title="Automatic Speech Recognition", # description = "Demo for Automatic Speech Recognition. Use microphone to record speech. Please press Record button. Initially it will take some time to load the model. The recognized text will appear in the output textbox", # ).launch() ###################################################### demo=gr.Interface( fn=sel_lng, inputs=[ #gr.Dropdown(["Hindi","Hindi-trans","Odiya","Odiya-trans","Kannada","Kannada-trans","Telugu","Telugu-trans","Bangala","Bangala-trans"],value="Hindi",label="Select Language"), gr.Dropdown([ # "Hindi","Hindi-trans", "Odia_model1","Odiya_trans_model1","Odia_model2","Odia_trans_model2"],label="Select Language"), # "Assamese-LM","Assamese-Model2", # "Punjabi_Model1","Punjabi_Model1_Trans","Punjabi_Model_aug","Punjabi_Model_aug_Trans"],value="Hindi",label="Select Language"), gr.Audio(sources=["microphone","upload"], type="filepath"), #gr.Audio(sources="upload", type="filepath"), #"state" ], outputs=[ "textbox" # #"state" ], allow_flagging="auto", #flagging_options=["Language error", "English transliteration error", "Other"], #flagging_callback=hf_writer, title="Automatic Speech Recognition", description = "Demo for Automatic Speech Recognition. Use microphone to record speech. Please press Record button. Initially it will take some time to load the model. The recognized text will appear in the output textbox", ).launch()