File size: 17,598 Bytes
541f2b9
 
eb43d71
541f2b9
 
 
 
 
 
 
 
 
 
 
 
 
f2e18d1
 
 
2f47bf1
f2e18d1
 
 
 
 
 
2f47bf1
 
541f2b9
8520752
 
541f2b9
af7d63c
541f2b9
0c92e5a
8520752
 
 
 
 
 
 
 
 
61f5a06
 
 
 
 
 
 
eea7953
 
61f5a06
 
 
8520752
541f2b9
 
 
 
 
 
 
 
 
 
 
 
 
124dbfa
541f2b9
eb43d71
541f2b9
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
eb43d71
541f2b9
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
f2e18d1
8d881b7
5bcaa28
2f47bf1
8d881b7
2f47bf1
eb43d71
 
 
f2e18d1
8d881b7
eb43d71
f2e18d1
 
 
 
 
 
 
8d881b7
 
 
 
f2e18d1
 
 
 
 
2f47bf1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
f2e18d1
 
 
 
 
 
2f47bf1
 
 
f2e18d1
2f47bf1
f2e18d1
 
 
 
 
61f5a06
 
2f47bf1
f2e18d1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5bcaa28
f2e18d1
5bcaa28
f2e18d1
 
 
 
 
 
 
 
 
 
 
 
8d881b7
eb43d71
541f2b9
 
 
 
 
 
 
 
 
8d881b7
541f2b9
 
 
8d881b7
541f2b9
 
 
 
 
 
 
 
 
 
 
8d881b7
541f2b9
 
 
8d881b7
541f2b9
 
 
 
 
 
 
 
 
 
 
 
 
 
eb43d71
541f2b9
 
 
 
 
 
8d881b7
541f2b9
 
 
 
eb43d71
2f47bf1
 
 
acdefab
2f47bf1
 
 
 
 
879bcf9
2f47bf1
 
 
 
 
acdefab
2f47bf1
 
 
 
 
 
 
 
 
eb43d71
 
acdefab
eb43d71
 
 
 
32de441
eb43d71
 
 
 
acdefab
eb43d71
 
 
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
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
from __future__ import unicode_literals
import youtube_dl
import yt_dlp
from pydub import AudioSegment
from pyannote.audio import Pipeline
import re
import whisper
import os
import ffmpeg
import subprocess
import gradio as gr
import traceback
import json
pipeline = Pipeline.from_pretrained("pyannote/speaker-diarization", use_auth_token="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")
from pydub.effects import speedup
import moviepy.editor as mp
import datetime
import torch
import pyannote.audio
from pyannote.audio.pipelines.speaker_verification import SpeechBrainPretrainedSpeakerEmbedding #PyannoteAudioPretrainedSpeakerEmbedding
from pyannote.audio import Audio
from pyannote.core import Segment
import wave
import contextlib
from sklearn.cluster import AgglomerativeClustering
import numpy as np
import json
from datetime import timedelta

from transformers import T5ForConditionalGeneration, T5Tokenizer

__FILES = set()
wispher_models = list(whisper._MODELS.keys())

def correct_grammar(input_text,num_return_sequences=1):
    torch_device = 'cuda' if torch.cuda.is_available() else 'cpu'
    tokenizer = T5Tokenizer.from_pretrained('deep-learning-analytics/GrammarCorrector')
    model = T5ForConditionalGeneration.from_pretrained('deep-learning-analytics/GrammarCorrector').to(torch_device)
    batch = tokenizer([input_text],truncation=True,padding='max_length',max_length=len(input_text), return_tensors="pt").to(torch_device)
    results = model.generate(**batch,max_length=len(input_text),num_beams=2, num_return_sequences=num_return_sequences, temperature=1.5)
    generated_sequences = []
    for generated_sequence_idx, generated_sequence in enumerate(results):
        text = tokenizer.decode(generated_sequence, clean_up_tokenization_spaces=True, skip_special_tokens=True)
        generated_sequences.append(text)
    generated_text = "".join(generated_sequences)
    _generated_text = ""
    for idx, _sentence in enumerate(generated_text.split('.'), 0):
        if not idx:
            _generated_text+=_sentence+'.'
        elif _sentence[:1]!=' ':
            _generated_text+=' '+_sentence+'.'
        elif _sentence[:1]=='':
            pass
        else:
            _generated_text+=_sentence+'.'
    return _generated_text

def CreateFile(filename):
    __FILES.add(filename)
    return filename

def RemoveFile(filename):
    if (os.path.isfile(filename)):
        os.remove(filename)

def RemoveAllFiles():
    for file in __FILES:
        if (os.path.isfile(file)):
            os.remove(file)
    
def Transcribe_V1(NumberOfSpeakers, SpeakerNames="", audio="temp_audio.wav"):
    SPEAKER_DICT = {}
    SPEAKERS = [speaker.strip() for speaker in SpeakerNames.split(',') if len(speaker)]
    
    def GetSpeaker(sp):
        speaker = sp
        if sp not in list(SPEAKER_DICT.keys()):
            if len(SPEAKERS):
                t = SPEAKERS.pop(0)
                SPEAKER_DICT[sp] = t
                speaker = SPEAKER_DICT[sp]
        else:
            speaker = SPEAKER_DICT[sp]
        return speaker
        
    def millisec(timeStr):
        spl = timeStr.split(":")
        s = (int)((int(spl[0]) * 60 * 60 + int(spl[1]) * 60 + float(spl[2]) )* 1000)
        return s
    
    def preprocess(audio):
        t1 = 0 * 1000 
        t2 = 20 * 60 * 1000
        newAudio = AudioSegment.from_wav(audio)
        a = newAudio[t1:t2]
        spacermilli = 2000
        spacer = AudioSegment.silent(duration=spacermilli)
        newAudio = spacer.append(a, crossfade=0)
        newAudio.export(audio, format="wav")
        return spacermilli, spacer
    
    def diarization(audio):
        as_audio = AudioSegment.from_wav(audio)
        DEMO_FILE = {'uri': 'blabal', 'audio': audio}
        if NumberOfSpeakers:
            dz = pipeline(DEMO_FILE, num_speakers=NumberOfSpeakers)  
        else:
            dz = pipeline(DEMO_FILE)  
        with open(CreateFile(f"diarization_{audio}.txt"), "w") as text_file:
            text_file.write(str(dz))
        dz = open(CreateFile(f"diarization_{audio}.txt")).read().splitlines()
        dzList = []
        for l in dz:
            start, end =  tuple(re.findall('[0-9]+:[0-9]+:[0-9]+\.[0-9]+', string=l))
            start = millisec(start)
            end = millisec(end)
            lex = GetSpeaker(re.findall('(SPEAKER_[0-9][0-9])', string=l)[0])
            dzList.append([start, end, lex])
        sounds = spacer
        segments = []
        dz = open(CreateFile(f"diarization_{audio}.txt")).read().splitlines()
        for l in dz:
            start, end =  tuple(re.findall('[0-9]+:[0-9]+:[0-9]+\.[0-9]+', string=l))
            start = millisec(start)
            end = millisec(end) 
            segments.append(len(sounds))
            sounds = sounds.append(as_audio[start:end], crossfade=0)
            sounds = sounds.append(spacer, crossfade=0)
        sounds.export(CreateFile(f"dz_{audio}.wav"), format="wav")
        return f"dz_{audio}.wav", dzList, segments
    
    def transcribe(dz_audio):
        model = whisper.load_model("medium")
        result = model.transcribe(dz_audio)
        # for _ in result['segments']:
        #     print(_['start'], _['end'], _['text'])
        captions = [[((caption["start"]*1000)), ((caption["end"]*1000)),  caption["text"]] for caption in result['segments']]
        conversation = []
        for i in range(len(segments)):
            idx = 0
            for idx in range(len(captions)):
                if captions[idx][0] >= (segments[i] - spacermilli):
                    break;
            
            while (idx < (len(captions))) and ((i == len(segments) - 1) or (captions[idx][1] < segments[i+1])):
                  c = captions[idx]  
                  start = dzList[i][0] + (c[0] -segments[i])
                  if start < 0: 
                      start = 0
                  idx += 1
                  if not len(conversation):
                      conversation.append([dzList[i][2], c[2]])
                  elif conversation[-1][0] == dzList[i][2]:
                      conversation[-1][1] +=  c[2]
                  else:
                      conversation.append([dzList[i][2], c[2]])
                  #print(f"[{dzList[i][2]}] {c[2]}")
        return conversation, ("".join([f"{speaker} --> {text}\n" for speaker, text in conversation]))

    spacermilli, spacer = preprocess(audio)
    dz_audio, dzList, segments = diarization(audio)
    conversation, t_text = transcribe(dz_audio)
    RemoveAllFiles()
    return (t_text, ({ "data": [{"speaker": speaker, "text": text} for speaker, text in conversation]}))


def Transcribe_V2(model, num_speakers, speaker_names, audio="temp_audio.wav"):
    model = whisper.load_model(model)
    # embedding_model = SpeechBrainPretrainedSpeakerEmbedding("speechbrain/spkrec-ecapa-voxceleb")
    
    embedding_model = SpeechBrainPretrainedSpeakerEmbedding(
        "speechbrain/spkrec-ecapa-voxceleb",
        device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
    )
    SPEAKER_DICT = {}
    default_speaker_names = ['A','B','C','D','E','F','G','H','I','J','K','L','M','N','O','P','Q','R','S','T','U','V','W','X','Y','Z']
    SPEAKERS = [speaker.strip() for speaker in speaker_names.split(',') if len(speaker)]
    def GetSpeaker(sp):
        speaker = sp
        if sp not in list(SPEAKER_DICT.keys()):
            if len(SPEAKERS):
                t = SPEAKERS.pop(0)
                SPEAKER_DICT[sp] = t
                speaker = SPEAKER_DICT[sp]
            elif len(default_speaker_names):
                t = default_speaker_names.pop(0)
                SPEAKER_DICT[sp] = t
                speaker = SPEAKER_DICT[sp]
        else:
            speaker = SPEAKER_DICT[sp]
        return speaker
    
    # audio = Audio()
    def diarization(audio):
        def millisec(timeStr):
            spl = timeStr.split(":")
            s = (int)((int(spl[0]) * 60 * 60 + int(spl[1]) * 60 + float(spl[2]) )* 1000)
            return s
        as_audio = AudioSegment.from_wav(audio)
        DEMO_FILE = {'uri': 'blabal', 'audio': audio}
        hparams = pipeline.parameters(instantiated=True)
        hparams["segmentation"]["min_duration_off"] -= 0.25
        pipeline.instantiate(hparams)
        if num_speakers:
            dz = pipeline(DEMO_FILE, num_speakers=num_speakers)  
        else:
            dz = pipeline(DEMO_FILE)  
        with open(CreateFile(f"diarization_{audio}.txt"), "w") as text_file:
            text_file.write(str(dz))
        dz = open(CreateFile(f"diarization_{audio}.txt")).read().splitlines()
        print(dz)
        dzList = []
        for l in dz:
            start, end =  tuple(re.findall('[0-9]+:[0-9]+:[0-9]+\.[0-9]+', string=l))
            start = millisec(start)
            end = millisec(end)
            lex = GetSpeaker(re.findall('(SPEAKER_[0-9][0-9])', string=l)[0])
            dzList.append([start, end, lex])
        return dzList
    
    def get_output(segments):
        # print(segments)
        conversation=[]
        for (i, segment) in enumerate(segments):
            # print(f"{i}, {segment["speaker"]}, {segments[i - 1]["speaker"]}, {}")
            if not len(conversation):
                conversation.append([str(timedelta(seconds=float(segment['start']))),str(timedelta(seconds=float(segment['end']))),GetSpeaker(segment["speaker"]), segment["text"].lstrip()])
            elif conversation[-1][2] == GetSpeaker(segment["speaker"]):
                conversation[-1][3] +=  segment["text"].lstrip()
            else:
                conversation.append([str(timedelta(seconds=float(segment['start']))),str(timedelta(seconds=float(segment['end']))),GetSpeaker(segment["speaker"]), segment["text"].lstrip()])
            # if i == 0 or segments[i - 1]["speaker"] != segment["speaker"]:
            #     if i != 0:
            #         conversation.append([GetSpeaker(segment["speaker"]), segment["text"][1:]]) # segment["speaker"] + ' ' + str(time(segment["start"])) + '\n\n'
            # conversation[-1][1] += segment["text"][1:]
        # return output
        for idx in range(len(conversation)):
            conversation[idx][3] = correct_grammar(conversation[idx][3])
        return ("".join([f"[{start}] - {speaker} \n{text}\n" for start, end, speaker, text in conversation])), ({ "data": [{"start": start, "end":end, "speaker": speaker, "text": text} for start, end, speaker, text in conversation]})

    def get_duration(path):
        with contextlib.closing(wave.open(path,'r')) as f:
            frames = f.getnframes()
            rate = f.getframerate()
        return frames / float(rate)

    def make_embeddings(path, segments, duration):
        embeddings = np.zeros(shape=(len(segments), 192))
        for i, segment in enumerate(segments):
            embeddings[i] = segment_embedding(path, segment, duration)
        return np.nan_to_num(embeddings)

    def segment_embedding(path, segment, duration):
        start = segment["start"]
        # Whisper overshoots the end timestamp in the last segment
        end = min(duration, segment["end"])
        clip = Segment(start, end)
        waveform, sample_rate = Audio().crop(path, clip)
        return embedding_model(waveform[None])

    def add_speaker_labels(segments, embeddings, num_speakers):
        clustering = AgglomerativeClustering(num_speakers).fit(embeddings)
        labels = clustering.labels_
        for i in range(len(segments)):
            segments[i]["speaker"] = 'SPEAKER ' + str(labels[i] + 1)

    def time(secs):
        return datetime.timedelta(seconds=round(secs))

    duration = get_duration(audio)
    if duration > 4 * 60 * 60:
        return "Audio duration too long"

    # print(json.dumps(diarization(audio)))
    result = model.transcribe(audio)
    # print(json.dumps(result))

    segments = result["segments"]

    num_speakers = min(max(round(num_speakers), 1), len(segments))
    if len(segments) == 1:
        segments[0]['speaker'] = 'SPEAKER 1'
    else:
        embeddings = make_embeddings(audio, segments, duration)
        add_speaker_labels(segments, embeddings, num_speakers)
    return get_output(segments)
    # return output

def AudioTranscribe(NumberOfSpeakers=None, SpeakerNames="", audio="", retries=5, model='base'):
    print(f"{NumberOfSpeakers}, {SpeakerNames}, {retries}")
    if retries:
        # subprocess.call(['ffmpeg', '-i', audio,'temp_audio.wav'])
        try:
            subprocess.call(['ffmpeg', '-i', audio,'temp_audio.wav'])
        except Exception as ex:
            traceback.print_exc()
            return AudioTranscribe(NumberOfSpeakers, SpeakerNames, audio, retries-1)
        if not (os.path.isfile("temp_audio.wav")):
            return AudioTranscribe(NumberOfSpeakers, SpeakerNames, audio, retries-1)
        return Transcribe_V2(model, NumberOfSpeakers, SpeakerNames)
    else:
        raise gr.Error("There is some issue ith Audio Transcriber. Please try again later!")

def VideoTranscribe(NumberOfSpeakers=None, SpeakerNames="", video="", retries=5, model='base'):
    if retries:
        try:
            clip = mp.VideoFileClip(video)
            clip.audio.write_audiofile("temp_audio.wav")
            # command = f"ffmpeg -i {video} -ab 160k -ac 2 -ar 44100 -vn temp_audio.wav"
            # subprocess.call(command, shell=True)
        except Exception as ex:
            traceback.print_exc()
            return VideoTranscribe(NumberOfSpeakers, SpeakerNames, video, retries-1)
        if not (os.path.isfile("temp_audio.wav")):
            return VideoTranscribe(NumberOfSpeakers, SpeakerNames, video, retries-1)
        return Transcribe_V2(model, NumberOfSpeakers, SpeakerNames)
    else:
        raise gr.Error("There is some issue ith Video Transcriber. Please try again later!")

def YoutubeTranscribe(NumberOfSpeakers=None, SpeakerNames="", URL="", retries = 5, model='base'):
    if retries:
        if "youtu" not in URL.lower():
            raise gr.Error(f"{URL} is not a valid youtube URL.")
        else:
            RemoveFile("temp_audio.wav")
            ydl_opts = {
                'format': 'bestaudio/best',
                'outtmpl': 'temp_audio.%(ext)s',
                'postprocessors': [{
                    'key': 'FFmpegExtractAudio',
                    'preferredcodec': 'wav',
                }],
            }
            try:
              with yt_dlp.YoutubeDL(ydl_opts) as ydl:
                  ydl.download([URL])
            except:
                return YoutubeTranscribe(NumberOfSpeakers, SpeakerNames, URL, retries-1)
            stream = ffmpeg.input('temp_audio.m4a')
            stream = ffmpeg.output(stream, 'temp_audio.wav')
            RemoveFile("temp_audio.m4a")
            return Transcribe_V2(model, NumberOfSpeakers, SpeakerNames)
    else:
        raise gr.Error(f"Unable to get video from {URL}")
 

with gr.Blocks() as yav_ui:
    with gr.Row():
        with gr.Column():
            with gr.Tab("Youtube", id=1):
                ysz = gr.Dropdown(label="Model Size", choices=wispher_models , value='base')
                yinput_nos = gr.Number(label="Number of Speakers", placeholder="2")
                yinput_sn = gr.Textbox(label="Name of the Speakers (ordered by the time they speak and separated by comma)", placeholder="If Speaker 1 is first to speak followed by Speaker 2 then -> Speaker 1, Speaker 2")
                yinput = gr.Textbox(label="Youtube Link", placeholder="https://www.youtube.com/watch?v=GECcjrYHH8w")
                ybutton_transcribe = gr.Button("Transcribe", show_progress=True, scroll_to_output=True)
            with gr.Tab("Video", id=2):
                vsz = gr.Dropdown(label="Model Size", choices=wispher_models, value='base')
                vinput_nos = gr.Number(label="Number of Speakers", placeholder="2")
                vinput_sn = gr.Textbox(label="Name of the Speakers (ordered by the time they speak and separated by comma)", placeholder="If Speaker 1 is first to speak followed by Speaker 2 then -> Speaker 1, Speaker 2")
                vinput = gr.Video(label="Video")
                vbutton_transcribe = gr.Button("Transcribe", show_progress=True, scroll_to_output=True)
            with gr.Tab("Audio", id=3):
                asz = gr.Dropdown(label="Model Size", choices=wispher_models , value='base')
                ainput_nos = gr.Number(label="Number of Speakers", placeholder="2")
                ainput_sn = gr.Textbox(label="Name of the Speakers (ordered by the time they speak and separated by comma)", placeholder="If Speaker 1 is first to speak followed by Speaker 2 then -> Speaker 1, Speaker 2")
                ainput = gr.Audio(label="Audio", type="filepath")
                abutton_transcribe = gr.Button("Transcribe", show_progress=True, scroll_to_output=True)
        with gr.Column():
            with gr.Tab("Text"):
                output_textbox = gr.Textbox(label="Transcribed Text", lines=15)
            with gr.Tab("JSON"):
                output_json = gr.JSON(label="Transcribed JSON")
    ybutton_transcribe.click(
                fn=YoutubeTranscribe,
                inputs=[yinput_nos,yinput_sn,yinput, ysz],
                outputs=[output_textbox,output_json]
            )
    abutton_transcribe.click(
                fn=AudioTranscribe,
                inputs=[ainput_nos,ainput_sn,ainput, asz],
                outputs=[output_textbox,output_json]
            )
    vbutton_transcribe.click(
                fn=VideoTranscribe,
                inputs=[vinput_nos,vinput_sn,vinput, vsz],
                outputs=[output_textbox,output_json]
            )
yav_ui.launch(debug=True)