Spaces:

chrisjay
/

afro-speech

Build error

App Files Files Community

chrisjay commited on May 10, 2022

Commit

81e77f2

1 Parent(s): ddb2b33

updates to afro speech

Browse files

Files changed (14) hide show

app.css +38 -0
app.py +160 -85
number/0.jpg +0 -0
number/1.jpg +0 -0
number/2.jpg +0 -0
number/3.jpg +0 -0
number/4.jpg +0 -0
number/5.jpg +0 -0
number/6.jpg +0 -0
number/7.jpg +0 -0
number/8.jpg +0 -0
number/9.jpg +0 -0
number/best.gif +3 -0
utils.py +44 -0

app.css ADDED Viewed

	@@ -0,0 +1,38 @@

+.infoPoint h1 {
+    font-size: 30px;
+    text-decoration: bold;
+    }
+a {
+    text-decoration: underline;
+    color: #1f3b54 ;
+}
+.finished {
+    color:rgb(9, 102, 169);
+     font-size:13px
+}
+table {
+    margin: 25px 0;
+    font-size: 0.9em;
+    font-family: sans-serif;
+    min-width: 400px;
+    max-width: 400px;
+    box-shadow: 0 0 20px rgba(0, 0, 0, 0.15);
+}
+table th,
+table td {
+    padding: 12px 15px;
+}
+tr {
+text-align: left;
+}
+thead tr {
+text-align: left;
+}

app.py CHANGED Viewed

@@ -1,144 +1,219 @@
 import os
 import csv
 import pandas as pd
 import gradio as gr
-import huggingface_hub
 import scipy.io.wavfile as wavf
-from huggingface_hub import Repository
 HF_TOKEN = os.environ.get("HF_TOKEN")
 DATASET_REPO_URL = "https://huggingface.co/datasets/chrisjay/crowd-speech-africa"
-DATA_FILENAME = "data.csv"
-AUDIO_PATH = os.path.join("data",'wav')
-DATA_FILE = os.path.join("data", DATA_FILENAME)
-# Get a dropdown of all African languages
-DEFAULT_LANGS = {'Igbo':'ibo','Yoruba':'yor','Hausa':'hau'}
 repo = Repository(
     local_dir="data", clone_from=DATASET_REPO_URL, use_auth_token=HF_TOKEN
 )
-repo.git_pull()
-os.makedirs(AUDIO_PATH,exist_ok=True)
-def push_record():
-    # Push the wav to a folder and reference the location
-    commit_url = repo.push_to_hub()
-    output = f'Recordings successfully pushed!'
-    output_string = "<html> <body> <div class='output' style='color:green; font-size:13px'>"+output+"</div> </body> </html>"
-    return output_string
-def save_record(language,text,record):
     # Save text and its corresponding record to flag
-    if language!=None and language!='Choose language' and record is not None:
-        lang_id = DEFAULT_LANGS[language]
-        text =text.strip()
-        # Write audio to file
-        audio_output_filename = os.path.join(AUDIO_PATH,f'{len(os.listdir(AUDIO_PATH))}.wav')
-        import pdb; pdb.set_trace()
-        wavf.write(audio_output_filename,record[0],record[1])
-        if not os.path.exists(DATA_FILE):
-            # Add header (necessary for listen)
-            with open(DATA_FILE, "a") as csvfile:
-                writer = csv.DictWriter(csvfile, fieldnames=["language", "audio","text"])
-                writer.writeheader()
-        with open(DATA_FILE, "a") as csvfile:
-            writer = csv.DictWriter(csvfile, fieldnames=["language", "audio","text"])
-            writer.writerow(
-                {"language": lang_id, "audio": audio_output_filename,"text": text}
-            )
-        output = f'Recording successfully saved! Click `Push` when you are done to send your recordings to the repo.'
-    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
 def display_records():
-    df = pd.read_csv(DATA_FILE)
-    langs=df['language'].values
-    df['audio'] = df['audio'].apply(lambda x: x.replace('data/wav/','https://huggingface.co/datasets/chrisjay/crowd-speech-africa/resolve/main/wav/'))
-    audios = df['audio'].values
-    texts=df['text'].values
-    html = """<div>
-                <table>
                     <tr>
                         <th>language</th>
                         <th>audio</th>
                         <th>text</th>
                     </tr>"""
-    for lang, audio, text in zip(langs,audios,texts):
         html+= f"""<tr>
                     <td>{lang}</td>
                     <td><audio controls><source src="{audio}" type="audio/wav"> </audio></td>
                     <td>{text}</td>
                 </tr>"""
     html+="</table></div>"
     return html
-title = 'African Crowdsource Speech'
-description = 'A platform to contribute to your African language by recording your voice'
-markdown = """# Africa Crowdsource Speech
-### a platform to contribute to your African language by recording your voice"""
-GENDER = ['Choose Gender','Male','Female','Other','Prefer not to say']
-NUMBERS = [i for i in range(21)]
 # Interface design begins
-block = gr.Blocks()
 with block:
     gr.Markdown(markdown)
     with gr.Tabs():
         with gr.TabItem('Record'):
-            #with gr.Row():
-            language = gr.inputs.Dropdown(choices = list(DEFAULT_LANGS.keys()),label="Choose language",default="Choose language")
             with gr.Row():
-                name = gr.inputs.Textbox(placeholder='e.g John Doe',label="Your name")
-                gender = gr.inputs.Dropdown(choices=GENDER, type="value", default=None, label="Gender")
-                accent = gr.inputs.Textbox(label="Accent (if any)")
-            number = gr.inputs.Radio(choices=NUMBERS, type="value", default=None, label="Choose your number")
-            with gr.Row():
-                text = gr.inputs.Textbox(placeholder='e.g `one` is `otu` in Igbo or `ọkan` in Yoruba',label="Number in your language")
-                record = gr.inputs.Audio(source="microphone",label='Record your voice')
             output_result = gr.outputs.HTML()
-            with gr.Row():
-                save = gr.Button("Save")
-                push = gr.Button('Submit')
-            save.click(save_record, inputs=[language,text,record],outputs=output_result)
-            push.click(push_record, inputs=[],outputs=output_result)
         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>.")
-            listen = gr.Button("Listen")
-            listen.click(display_records,inputs=[],outputs=gr.outputs.HTML() )
 block.launch()

+from email.policy import default
 import os
 import csv
+import random
 import pandas as pd
 import gradio as gr
+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>"""
 # 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="Number in your language")
+            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)
 block.launch()

number/0.jpg ADDED Viewed

number/1.jpg ADDED Viewed

number/2.jpg ADDED Viewed

number/3.jpg ADDED Viewed

number/4.jpg ADDED Viewed

number/5.jpg ADDED Viewed

number/6.jpg ADDED Viewed

number/7.jpg ADDED Viewed

number/8.jpg ADDED Viewed

number/9.jpg ADDED Viewed

number/best.gif ADDED Viewed

Git LFS Details

SHA256: 7cd04273ca701f7e8b6ae01e55a6185ed49dbe0527a23fb7443dccfae54cbcb7
Pointer size: 132 Bytes
Size of remote file: 1.36 MB

utils.py ADDED Viewed

	@@ -0,0 +1,44 @@

+import json
+import hashlib
+import random
+import string
+def get_unique_name():
+    return ''.join([random.choice(string.ascii_letters
+            + string.digits) for n in range(32)])
+def read_json_lines(file):
+    with open(file,'r',encoding="utf8") as f:
+        lines = f.readlines()
+        data=[]
+        for l in lines:
+            data.append(json.loads(l))
+        return data
+def json_dump(thing):
+    return json.dumps(thing,
+                        ensure_ascii=False,
+                        sort_keys=True,
+                        indent=None,
+                        separators=(',', ':'))
+def get_hash(thing): # stable-hashing
+    return str(hashlib.md5(json_dump(thing).encode('utf-8')).hexdigest())
+def dump_json(thing,file):
+    with open(file,'w+',encoding="utf8") as f:
+        json.dump(thing,f)
+def read_json_lines(file):
+    with open(file,'r',encoding="utf8") as f:
+        lines = f.readlines()
+        data=[]
+        for l in lines:
+            data.append(json.loads(l))
+        return data