import os import csv 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") DATASET_REPO_URL = "https://huggingface.co/datasets/chrisjay/crowd-speech-africa" #DATA_FILENAME = "data.csv" REPOSITORY_DIR = "data" LOCAL_DIR = 'data_local' #DATA_FILE = os.path.join("data", DATA_FILENAME) os.makedirs(LOCAL_DIR,exist_ok=True) # Get a dropdown of all African languages #DEFAULT_LANGS = {'Igbo':'ibo','Yoruba':'yor','Hausa':'hau'} GENDER = ['Choose Gender','Male','Female','Other','Prefer not to say'] NUMBERS = [i for i in range(21)] #------------------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() def save_record(language,text,record,number,age,gender,accent,number_history): 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 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':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!' 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 = "

"+output+"

" # return output_string, next image and state return output_string 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] html = """

""" for lang, audio, text in zip(langs,audios,texts): html+= f"""""" html+="

language	audio	text
{lang}		{text}

" return html NUMBER_DIR = './number' number_files = [f.name for f in os.scandir(NUMBER_DIR)] NUMBERS = [{'image':os.path.join(NUMBER_DIR,f),'number':int(f.split('.')[0])} for f in number_files] markdown = """

Africa Crowdsource Speech

This is 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 = 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") #number = gr.inputs.Radio(choices=NUMBERS, type="value", default=None, label="Choose your number") 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.inputs.State() save = gr.Button("Submit") save.click(save_record, inputs=[language,text,record,number,age,gender,accent,state],outputs=[output_result,number,state]) with gr.TabItem('Listen') as listen_tab: gr.Markdown("Listen to the recordings contributed. You can find them here.") listen = gr.Button("Listen") listen.click(display_records,inputs=[],outputs=gr.outputs.HTML() ) block.launch()