Spaces:
Build error
Build error
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 | |
from IPython.display import Audio | |
from IPython.core.display import display | |
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': | |
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') | |
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.' | |
else: | |
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) | |
#df['audio'] = df['audio'].apply(lambda x: display(Audio(x,autoplay=True))) | |
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 | |
#return langs[0],audios[0],texts[0] | |
#return df | |
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""" | |
# 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="language",default="Choose language") | |
text = gr.inputs.Textbox(placeholder='Write your text',label="text to record") | |
record = gr.inputs.Audio(source="microphone",label='Record your voice') | |
output_result = gr.outputs.HTML() | |
save = gr.Button("Save") | |
push = gr.Button('Push') | |
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 [here](https://huggingface.co/datasets/chrisjay/crowd-speech-africa).") | |
listen = gr.Button("Listen") | |
listen.click(display_records,inputs=[],outputs=gr.outputs.HTML() ) | |
block.launch(enable_queue=True) |