File size: 7,587 Bytes
8e73e42
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
bf4b781
8e73e42
bf4b781
 
 
8e73e42
 
 
 
 
 
 
 
 
 
 
 
 
bf4b781
8e73e42
bf4b781
 
 
8e73e42
 
 
 
 
 
 
 
 
 
 
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
import torch

import gradio as gr
import yt_dlp as youtube_dl
import numpy as np
from datasets import Dataset, Audio
from scipy.io import wavfile

from transformers import pipeline
from transformers.pipelines.audio_utils import ffmpeg_read

import tempfile
import os
import time


MODEL_NAME = "openai/whisper-large-v3"
BATCH_SIZE = 8
FILE_LIMIT_MB = 1000
YT_LENGTH_LIMIT_S = 3600  # limit to 1 hour YouTube files

device = 0 if torch.cuda.is_available() else "cpu"

pipe = pipeline(
    task="automatic-speech-recognition",
    model=MODEL_NAME,
    chunk_length_s=30,
    device=device,
)


def transcribe(inputs_path, task, dataset_name, oauth_token: gr.OAuthToken):
    if inputs_path is None:
        raise gr.Error("No audio file submitted! Please upload or record an audio file before submitting your request.")
    
    sampling_rate, inputs = wavfile.read(inputs_path) 

    out = pipe(inputs_path, batch_size=BATCH_SIZE, generate_kwargs={"task": task}, return_timestamps=True)
    
    text = out["text"]
    
    chunks = naive_postprocess_whisper_chunks(out["chunks"])
        
    transcripts = []
    audios = []
    with tempfile.TemporaryDirectory() as tmpdirname:
        for i,chunk in enumerate(chunks):
            begin, end = chunk["timestamp"]
            begin, end = int(begin*sampling_rate), int(end*sampling_rate)
            # TODO: make sure 1D or 2D?
            arr = inputs[begin:end]
            path = os.path.join(tmpdirname, f"{i}.wav")
            wavfile.write(path, sampling_rate,  arr)
            audios.append(path)
            transcripts.append(chunk["text"])
            
        dataset = Dataset.from_dict({"audio": audios, "transcript": transcripts}).cast_column("audio", Audio())
        
    
        dataset.push_to_hub(dataset_name, token=oauth_token)
        
    return  text


def _return_yt_html_embed(yt_url):
    video_id = yt_url.split("?v=")[-1]
    HTML_str = (
        f'<center> <iframe width="500" height="320" src="https://www.youtube.com/embed/{video_id}"> </iframe>'
        " </center>"
    )
    return HTML_str

def download_yt_audio(yt_url, filename):
    info_loader = youtube_dl.YoutubeDL()
    
    try:
        info = info_loader.extract_info(yt_url, download=False)
    except youtube_dl.utils.DownloadError as err:
        raise gr.Error(str(err))
    
    file_length = info["duration_string"]
    file_h_m_s = file_length.split(":")
    file_h_m_s = [int(sub_length) for sub_length in file_h_m_s]
    
    if len(file_h_m_s) == 1:
        file_h_m_s.insert(0, 0)
    if len(file_h_m_s) == 2:
        file_h_m_s.insert(0, 0)
    file_length_s = file_h_m_s[0] * 3600 + file_h_m_s[1] * 60 + file_h_m_s[2]
    
    if file_length_s > YT_LENGTH_LIMIT_S:
        yt_length_limit_hms = time.strftime("%HH:%MM:%SS", time.gmtime(YT_LENGTH_LIMIT_S))
        file_length_hms = time.strftime("%HH:%MM:%SS", time.gmtime(file_length_s))
        raise gr.Error(f"Maximum YouTube length is {yt_length_limit_hms}, got {file_length_hms} YouTube video.")
    
    ydl_opts = {"outtmpl": filename, "format": "worstvideo[ext=mp4]+bestaudio[ext=m4a]/best[ext=mp4]/best"}
    
    with youtube_dl.YoutubeDL(ydl_opts) as ydl:
        try:
            ydl.download([yt_url])
        except youtube_dl.utils.ExtractorError as err:
            raise gr.Error(str(err))


def yt_transcribe(yt_url, task, dataset_name, oauth_token: gr.OAuthToken, max_filesize=75.0, dataset_sampling_rate = 24000):
    html_embed_str = _return_yt_html_embed(yt_url)

    with tempfile.TemporaryDirectory() as tmpdirname:
        filepath = os.path.join(tmpdirname, "video.mp4")
        download_yt_audio(yt_url, filepath)
        with open(filepath, "rb") as f:
            inputs_path = f.read()

    inputs = ffmpeg_read(inputs_path, pipe.feature_extractor.sampling_rate)
    inputs = {"array": inputs, "sampling_rate": pipe.feature_extractor.sampling_rate}

    out = pipe(inputs, batch_size=BATCH_SIZE, generate_kwargs={"task": task}, return_timestamps=True)
    
    text = out["text"]
    
    chunks = naive_postprocess_whisper_chunks(out["chunks"])
    
    inputs = ffmpeg_read(inputs_path, dataset_sampling_rate)
    
    transcripts = []
    audios = []
    with tempfile.TemporaryDirectory() as tmpdirname:
        for i,chunk in enumerate(chunks):
            begin, end = chunk["timestamp"]
            begin, end = int(begin*dataset_sampling_rate), int(end*dataset_sampling_rate)
            # TODO: make sure 1D or 2D?
            arr = inputs[begin:end]
            path = os.path.join(tmpdirname, f"{i}.wav")
            wavfile.write(path, dataset_sampling_rate,  arr)
            audios.append(path)
            transcripts.append(chunk["text"])
            
        dataset = Dataset.from_dict({"audio": audios, "transcript": transcripts}).cast_column("audio", Audio())
        
    
        dataset.push_to_hub(dataset_name, token=oauth_token)
        

    return html_embed_str, text


def naive_postprocess_whisper_chunks(chunks, stop_chars = ".!:;?", min_duration = 5):
    new_chunks = []
    
    while chunks:
        current_chunk = chunks.pop(0)
        begin, end = current_chunk["timestamp"]
        text = current_chunk["text"]
        
        while chunks and (text[-1] not in stop_chars or (end-begin<min_duration)):
            ch = chunks.pop(0)
            end = ch["timestamp"][1]
            text = "".join([text, ch["text"]])
            
        new_chunks.append({
            "text": text.strip(),
            "timestamp": (begin, end),
        })
        print(f"LENGTH CHUNK #{len(new_chunks)}: {end-begin}s")
            
    return new_chunks
    
    
    
    
    

demo = gr.Blocks()

mf_transcribe = gr.Interface(
    fn=transcribe,
    inputs=[
        gr.Audio(type="filepath"),
        gr.Radio(["transcribe", "translate"], label="Task", value="transcribe"),
        gr.Textbox(lines=1, placeholder="Place your new dataset name here", label="Dataset name"),
    ],
    outputs="text",
    theme="huggingface",
    title="Create your own TTS dataset using your own recordings",
    description=(
        "This demo allows use to create a text-to-speech dataset from an input audio snippet and push it to hub to keep track of it."
        f"Demo uses the checkpoint [{MODEL_NAME}](https://huggingface.co/{MODEL_NAME}) and 🤗 Transformers to automatically transcribe audio files"
        " of arbitrary length. It then merge chunks of audio and push it to the hub."
    ),
    allow_flagging="never",
)

yt_transcribe = gr.Interface(
    fn=yt_transcribe,
    inputs=[
        gr.Textbox(lines=1, placeholder="Paste the URL to a YouTube video here", label="YouTube URL"),
        gr.Radio(["transcribe", "translate"], label="Task", value="transcribe"),
        gr.Textbox(lines=1, placeholder="Place your new dataset name here", label="Dataset name"),
    ],
    outputs=["html", "text"],
    theme="huggingface",
    title="Create your own TTS dataset using Youtube",
    description=(
        "This demo allows use to create a text-to-speech dataset from an input audio snippet and push it to hub to keep track of it."
        f"Demo uses the checkpoint [{MODEL_NAME}](https://huggingface.co/{MODEL_NAME}) and 🤗 Transformers to automatically transcribe audio files"
        " of arbitrary length. It then merge chunks of audio and push it to the hub."
    ),
    allow_flagging="never",
)

with demo:
    with gr.Row():
        gr.LoginButton()
        gr.LogoutButton()
    gr.TabbedInterface([mf_transcribe, yt_transcribe], ["Microphone or Audio file", "YouTube"])

demo.launch(debug=True)