app.py

import os
import csv
import random
import pandas as pd
import numpy as np
import gradio as gr
from collections import Counter 
from utils import *
import matplotlib.pyplot as plt
import scipy.io.wavfile as wavf
from huggingface_hub import Repository, upload_file



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


GREETINGS_DIR = './greetings'
greeting_files = [f.name for f in os.scandir(GREETINGS_DIR)]


DATASET_REPO_URL = "https://huggingface.co/datasets/meyabase/crowd-oshiwambo-speech-greetings"
REPOSITORY_DIR = "data"
LOCAL_DIR = 'data_local'
os.makedirs(LOCAL_DIR,exist_ok=True)


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

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

languages = ["oshindonga", "oshikwanyama"]
language_id = ["ng","kj"]


#------------------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,record,greeting,gender,accent,greeting_history,current_greeting,done_recording):
    # set default
    greeting_history = greeting_history if greeting_history is not None else [0]
    current_greeting = current_greeting if current_greeting is not None else 0 # 0 is the default greeting
    done_recording = done_recording if done_recording is not None else False
    #----

    # Save text and its corresponding record to flag
    speaker_metadata={}
    speaker_metadata['gender'] = gender if gender!=GENDER[0] else ''
    speaker_metadata['accent'] = accent if accent!='' else ''
    default_record = None
    if not done_recording:
        if language!=None and language!='Choose language' and record is not None and greeting is not None:  #
            language = language.lower()
            lang_id = language_id[languages.index(language)] 
    
            # 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,
                        'greeting':current_greeting, 
                        'frequency':record[0],
                        'gender': speaker_metadata['gender'],
                        'accent': speaker_metadata['accent'],
                    }
            
            dump_json(metadata,json_file_path)  
                
            # 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='meyabase/crowd-oshiwambo-speech-greetings',
                        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='meyabase/crowd-oshiwambo-speech-greetings',
                        repo_type='dataset',
                        token=HF_TOKEN
                    )        
            
            output = f'Recording successfully saved! On to the next one...'

            # Choose the next greeting 
            greeting_history.append(current_greeting)

            # check the language selected and choose the next greeting based on the images available
            if language=='oshindonga':
                greeting_choices = [greet for greet in [i for i in range(3)] if greet not in greeting_history]
                if greeting_choices!=[]:
                    next_greeting = random.choice(greeting_choices)
                    next_greeting_image = f'greetings/{language}/{next_greeting}.png'
                else:
                    done_recording=True
                    next_greeting = 0
                    next_greeting_image = 'greetings/best.gif'
                    output = "You have finished all recording! You can reload to start again."
                    
            elif language=='oshikwanyama':
                greeting_choices = [greet for greet in [i for i in range(3)] if greet not in greeting_history]
                if greeting_choices!=[]:
                    next_greeting = random.choice(greeting_choices)
                    next_greeting_image = f'greetings/{language}/{next_greeting}.png'
                else:
                    done_recording=True
                    next_greeting = 0
                    next_greeting_image = 'greetings/best.gif'
                    output = "You have finished all recording! You can reload to start again."
       
            output_string = "<html> <body> <div class='output' style='color:green; font-size:13px'>"+output+"</div> </body> </html>"
            return output_string,next_greeting_image,greeting_history,next_greeting,done_recording,default_record
        
        if greeting is None:
            output = "greeting 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, greeting,greeting_history,current_greeting,done_recording,default_record
    else:
                        
        # Stop submitting recording (best.gif is displaying)
        output = '🙌 You have finished all recording! Thank You. You can reload to start again.'
        output_string = "<div class='finished'>"+output+"</div>"
        next_greeting = 0 # the default greeting
        next_greeting_image = 'greetings/best.gif'    
        return output_string,next_greeting_image,greeting_history,next_greeting,done_recording,default_record


def get_metadata_json(path):
    try:
        return read_json_lines(path)[0]
    except Exception:
        return []    

def get_metadata_of_dataset():
    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/meyabase/crowd-oshiwambo-speech-greetings/resolve/main/data/') for a in audio_repo]
    metadata_all = [get_metadata_json(os.path.join(f,'metadata.jsonl')) for f in repo_recordings]
    metadata_all = [m for m in metadata_all if m!=[]]
    return metadata_all 

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/meyabase/crowd-oshiwambo-speech-greetings/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]
    greetings = [m['greeting'] 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>greeting</th>
                        <th>text</th>
                    </tr>""" 
    for lang, audio, text,greet_ in zip(langs,audios,texts,greetings):
        html+= f"""<tr>
                    <td>{lang}</td>
                    <td><audio controls><source src="{audio}" type="audio/wav"> </audio></td>
                    <td>{greet_}</td>
                    <td>{text}</td>
                </tr>"""      
    html+="</table></div>"
    return html
 
markdown = """<div style="text-align: center"><p style="font-size: 40px"> 🔊 <b>Oshiwambo Speech Greetings</b></p> <br>
This is a platform to contribute to your Oshiwambo greeting for the speech recognition task. <br> </div>"""

record_markdown = """
<br> Record greetings in your language and help us build a dataset for speech recognition in Oshiwambo. <br>
"""

# # Interface design begins
block = gr.Blocks(css=BLOCK_CSS)
with block:
    gr.Markdown(markdown)
    with gr.Tabs():
        
        with gr.TabItem('Record'):
            gr.Markdown(record_markdown)

            with gr.Row():
                language = gr.inputs.Dropdown(choices = sorted([lang_.title() for lang_ in list(languages)]), label="Choose language", default=languages[0].title())
                gender = gr.inputs.Dropdown(choices=GENDER, type="value", default=None, label="Gender (optional)")  
                accent = gr.inputs.Textbox(label="Accent (optional)", default='', placeholder="e.g. oshikwanyama, oshindonga, oshimbadja, oshingadjera, etc.") 

            # define a default greeting first for each language
            greeting = gr.Image(f'greetings/{languages[0].lower()}/0.png', image_mode="L")

            greeting_history = gr.Variable() # stores the history of greetings

            record = gr.Audio(source="microphone", label='Record your voice')
            output_result = gr.outputs.HTML()
            state = gr.Variable()
            current_greeting = gr.Variable()
            done_recording = gr.Variable() # Signifies when to stop submitting records even if `submit`` is clicked            
            save = gr.Button("Submit")
                
            save.click(save_record, inputs=[language,record,greeting,gender,accent,state,current_greeting,done_recording],outputs=[output_result,greeting,state,current_greeting,done_recording,record])

block.launch()