File size: 10,315 Bytes
81e77f2
45585f3
391decb
81e77f2
391decb
45585f3
74bf3d4
81e77f2
391decb
81e77f2
 
 
 
45585f3
391decb
45585f3
81e77f2
 
 
391decb
 
81e77f2
 
 
 
 
 
 
 
 
 
 
 
694e002
81e77f2
 
 
 
b829e6a
391decb
 
 
81e77f2
391decb
81e77f2
 
391decb
 
 
81e77f2
 
391decb
45585f3
81e77f2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
391decb
81e77f2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
391decb
 
81e77f2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b1ea4b4
 
 
81e77f2
b1ea4b4
 
81e77f2
b1ea4b4
 
 
81e77f2
b1ea4b4
 
 
 
893eec1
391decb
81e77f2
391decb
 
 
81e77f2
 
45585f3
 
74bf3d4
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
45585f3
81e77f2
fb86baf
 
391decb
 
 
52064f2
81e77f2
 
 
 
 
 
74bf3d4
81e77f2
52064f2
 
81e77f2
 
 
 
 
391decb
81e77f2
391decb
 
b5ee8ce
81e77f2
 
 
 
 
 
 
 
74bf3d4
45585f3
52064f2
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
from email.policy import default
import os
import csv
import random
import pandas as pd
import gradio as gr
from article import ARTICLE
from utils import *
import scipy.io.wavfile as wavf
from huggingface_hub import Repository, upload_file




HF_TOKEN = os.environ.get("HF_TOKEN")

NUMBER_DIR = './number' 
number_files = [f.name for f in os.scandir(NUMBER_DIR)]


DATASET_REPO_URL = "https://huggingface.co/datasets/chrisjay/crowd-speech-africa"
REPOSITORY_DIR = "data"
LOCAL_DIR = 'data_local'
os.makedirs(LOCAL_DIR,exist_ok=True)
#DEFAULT_LANGS = {'Igbo':'ibo','Yoruba':'yor','Hausa':'hau'}

GENDER = ['Choose Gender','Male','Female','Other','Prefer not to say']


#------------------Work on Languages--------------------
DEFAULT_LANGS =   {}
languages = read_json_lines('clean_languages.json')
languages_lower=[l for l in languages]

_ = [DEFAULT_LANGS.update({l['full'].lower():l['id'].lower()}) for l in languages_lower]
#_ = [DEFAULT_LANGS.update({l_other.lower():[l['id'].lower()]}) for l in languages_lower for l_other in l['others'] if l_other.lower()!=l['full'].lower()]

#------------------Work on Languages--------------------

repo = Repository(
    local_dir="data", clone_from=DATASET_REPO_URL, use_auth_token=HF_TOKEN
)
#repo.git_pull()

with open('app.css','r') as f:
    BLOCK_CSS = f.read() 



def save_record(language,text,record,number,age,gender,accent,number_history,current_number,done_recording):
    number_history = number_history or [0]

    # Save text and its corresponding record to flag
    speaker_metadata={}
    speaker_metadata['gender'] = gender if gender!=GENDER[0] else ''
    speaker_metadata['age'] = age if age !='' else ''
    speaker_metadata['accent'] = accent if accent!='' else ''
    if not done_recording:
        if language!=None and language!='Choose language' and record is not None and number is not None:
            language = language.lower()
            lang_id = DEFAULT_LANGS[language]
            text =text.strip()
            
            # Write audio to file
            audio_name = get_unique_name()
            SAVE_FILE_DIR = os.path.join(LOCAL_DIR,audio_name)
            os.makedirs(SAVE_FILE_DIR,exist_ok=True)
            audio_output_filename = os.path.join(SAVE_FILE_DIR,'audio.wav')
            wavf.write(audio_output_filename,record[0],record[1])

            # Write metadata.json to file
            json_file_path = os.path.join(SAVE_FILE_DIR,'metadata.jsonl')
            metadata= {'id':audio_name,'file_name':'audio.wav',
                        'language_name':language,'language_id':lang_id,
                        'number':current_number, 'text':text,'frequency':record[0],
                        'age': speaker_metadata['age'],'gender': speaker_metadata['gender'],
                        'accent': speaker_metadata['accent']
                        }
            
            dump_json(metadata,json_file_path)  
                
            # Simply upload the audio file and metadata using the hub's upload_file
            # Upload the audio
            repo_audio_path = os.path.join(REPOSITORY_DIR,os.path.join(audio_name,'audio.wav'))
            
            _ = upload_file(path_or_fileobj = audio_output_filename,
                        path_in_repo =repo_audio_path,
                        repo_id='chrisjay/crowd-speech-africa',
                        repo_type='dataset',
                        token=HF_TOKEN
                    ) 

            # Upload the metadata
            repo_json_path = os.path.join(REPOSITORY_DIR,os.path.join(audio_name,'metadata.jsonl'))
            _ = upload_file(path_or_fileobj = json_file_path,
                        path_in_repo =repo_json_path,
                        repo_id='chrisjay/crowd-speech-africa',
                        repo_type='dataset',
                        token=HF_TOKEN
                    )        
            
            output = f'Recording successfully saved!'

            # Choose the next number 
            number_history.append(current_number)
            number_choices = [num for num in [i for i in range(10)] if num not in number_history]
            if number_choices!=[]:
                next_number = random.choice(number_choices)

                next_number_image = f'number/{next_number}.jpg'
            else:
                done_recording=True
                next_number = 0 # the default number
                next_number_image = f'number/best.gif'    
            output_string = "<html> <body> <div class='output' style='color:green; font-size:13px'>"+output+"</div> </body> </html>"
            return output_string,next_number_image,number_history,next_number,done_recording
        
        if number is None:
            output = "Number must be specified!"
        if record is None:
            output="No recording found!"
        if language is None or language=='Choose language':
            output = 'Language must be specified!'     
        output_string = "<html> <body> <div class='output' style='color:green; font-size:13px'>"+output+"</div> </body> </html>"

        # return output_string, previous image and state 
        return output_string, number,number_history,current_number,done_recording
    else:
        # Stop submitting recording (best.gif is displaying)
        output = '🙌 You have finished all recording! Thank You. You can reload to start again (maybe in another language).'
        output_string = "<div class='finished'>"+output+"</div>"
        next_number = 0 # the default number
        next_number_image = f'number/best.gif'    
        return output_string,next_number_image,number_history,next_number,done_recording

def display_records():
    repo.git_pull()
    REPOSITORY_DATA_DIR = os.path.join(REPOSITORY_DIR,'data')
    repo_recordings = [os.path.join(REPOSITORY_DATA_DIR,f.name) for f in os.scandir(REPOSITORY_DATA_DIR)] if os.path.isdir(REPOSITORY_DATA_DIR) else []

    audio_repo = [os.path.join(f,'audio.wav') for f in repo_recordings]
    audio_repo = [a.replace('data/data/','https://huggingface.co/datasets/chrisjay/crowd-speech-africa/resolve/main/data/') for a in audio_repo]
    metadata_repo = [read_json_lines(os.path.join(f,'metadata.jsonl'))[0] for f in repo_recordings]
    audios_all =  audio_repo
    metadata_all = metadata_repo

 
    langs=[m['language_name'] for m in metadata_all]
    audios = [a for a in audios_all]
    texts = [m['text'] for m in metadata_all]
    numbers = [m['number'] for m in metadata_all]


    html = f"""<div class="infoPoint">
                <h1> Hooray! We have collected {len(metadata_all)} samples!</h1> 
                <table style="width:100%; text-align:center">
                    <tr>
                        <th>language</th>
                        <th>audio</th>
                        <th>number</th>
                        <th>text</th>
                    </tr>""" 
    for lang, audio, text,num_ in zip(langs,audios,texts,numbers):
        html+= f"""<tr>
                    <td>{lang}</td>
                    <td><audio controls><source src="{audio}" type="audio/wav"> </audio></td>
                    <td>{num_}</td>
                    <td>{text}</td>
                </tr>"""      
    html+="</table></div>"
    return html
 

# NUMBERS = [{'image':os.path.join(NUMBER_DIR,f),'number':int(f.split('.')[0])} for f in number_files] 



markdown = """<div style="text-align: center"><p style="font-size: 40px"> Africa Crowdsource Speech </p> <br>
This is a platform to contribute to your African language by recording your voice </div>"""


markdown="""
# 🌍 African Digits Recording Sprint

> This is an event to record numbers (0-9) in your African language.

1. Choose your African language
2. Fill in the speaker metadata (age, gender, accent). This is optional but important to build better speech models.
3. You will see the image of a number __(this is the number you will record)__.
4. Fill in the word of that **number** (optional)
5. Click record and say the number in your African language.
6. Click ‘Submit’. It will save your record and go to the next number. 
7. Repeat 1-7
8. Leave a ❤ in the Space, if you found it fun.
"""



# Interface design begins
block = gr.Blocks(css=BLOCK_CSS)
with block:
    gr.Markdown(markdown)
    with gr.Tabs():
        
        with gr.TabItem('Record'):
            with gr.Row():
                language = gr.inputs.Dropdown(choices = sorted([lang_.title() for lang_ in list(DEFAULT_LANGS.keys())]),label="Choose language",default="Choose language")
                age = gr.inputs.Textbox(placeholder='e.g. 21',label="Your age (optional)",default='')  
                gender = gr.inputs.Dropdown(choices=GENDER, type="value", default=None, label="Gender (optional)")  
                accent = gr.inputs.Textbox(label="Accent (optional)",default='')  

            number = gr.Image('number/0.jpg',image_mode="L") 
            text = gr.inputs.Textbox(placeholder='e.g. `one` is `otu` in Igbo or `ọkan` in Yoruba',label="How is the number called in your language (optional)")
            record = gr.inputs.Audio(source="microphone",label='Record your voice')
            
            output_result = gr.outputs.HTML()
            state = gr.Variable(default_value=[0])
            current_number = gr.Variable(default_value=0)
            done_recording = gr.Variable(default_value=False) # Signifies when to stop submitting records even if `submit`` is clicked            
            save = gr.Button("Submit")
                
            
            save.click(save_record, inputs=[language,text,record,number,age,gender,accent,state,current_number,done_recording],outputs=[output_result,number,state,current_number,done_recording])

        with gr.TabItem('Listen') as listen_tab:
            gr.Markdown("Listen to the recordings contributed. You can find them <a href='https://huggingface.co/datasets/chrisjay/crowd-speech-africa' target='blank'>here</a>.")
            display_html =  gr.HTML("""<div style="color: green">
                <p> ⌛ Please wait. Loading dataset... </p>
                </div>
               """)
            

            #listen = gr.Button("Listen")
            listen_tab.select(display_records,inputs=[],outputs=display_html)
    gr.Markdown(ARTICLE)        

block.launch()