app.py

import io
import os
import base64
import librosa
import numpy as np
from io import BytesIO
import streamlit as st
from pydub import AudioSegment
import matplotlib.pyplot as plt
from scipy.io.wavfile import write
from src.denoise import denoise
from myrecorder import recorder


SR = 16000
CONTAINER_HEIGHT = 340


def np_audio_to_bytesio(np_audio, np_audio_sr):
    _bytes = bytes()
    byte_io = io.BytesIO(_bytes)
    write(byte_io, np_audio_sr, np_audio)
    bytes_audio = byte_io.read()
    return bytes_audio


def autoplay_audio(audio: str):
    audio_base64 = base64.b64encode(audio).decode('utf-8')
    audio_tag = f'<audio autoplay="true" src="data:audio/wav;base64,{audio_base64}">'
    st.markdown(audio_tag, unsafe_allow_html=True)


def load_noisy_speech(root=os.path.join(os.getcwd(), 'noisy_speech')):
    noisy_speech_paths = {'EN':{}, 'JA': {}}
    noisy_speech_names = os.listdir(root)
    for name in noisy_speech_names:
        splt = name.split('_')
        lang, snr = splt[0].upper(), int(splt[1][:2])
        noisy_speech_paths[lang][snr] = os.path.join(root, name)
        
    en_keys = list(noisy_speech_paths['EN'].keys())
    en_keys.sort()
    en_keys.reverse()
    noisy_speech_paths['EN'] = {f'{key}dB': noisy_speech_paths['EN'][key] for key in en_keys}
    
    ja_keys = list(noisy_speech_paths['JA'].keys())
    ja_keys.sort()
    ja_keys.reverse()
    noisy_speech_paths['JA'] = {f'{key}dB': noisy_speech_paths['JA'][key] for key in ja_keys}
    
    return noisy_speech_paths


def load_wav(wav_path):
    wav_22k, sr = librosa.load(wav_path)
    wav_16k = librosa.resample(wav_22k, orig_sr=sr, target_sr=SR)
    return wav_22k, wav_16k


def wav_to_spec(wav, sr):
    if sr == 16000:
        wav = librosa.resample(wav, orig_sr=sr, target_sr=22050)
    spec = np.abs(librosa.stft(wav))
    spec = librosa.amplitude_to_db(spec, ref=np.max)
    return spec


def export_spec_to_buffer(spec):
    plt.clf()
    plt.rcParams['figure.figsize'] = (16, 3.6)
    plt.rc('axes', labelsize=15)
    plt.rc('xtick', labelsize=15)
    plt.rc('ytick', labelsize=15)
    librosa.display.specshow(spec, y_axis='linear', x_axis='time')
    img_buffer = BytesIO()
    img_buffer.truncate(0)  # Remove all contents
    img_buffer.seek(0)  # Reset the pointer to the start
    plt.savefig(img_buffer, format='JPEG', bbox_inches='tight', pad_inches=0)
    plt.close('all')
    return img_buffer


def process_recorded_wav_bytes(wav_bytes, sr):
    file = BytesIO(wav_bytes)
    audio = AudioSegment.from_file(file=file, format='wav')
    audio = audio.set_sample_width(2)
    audio = audio.set_channels(1)
    audio_22k = audio.set_frame_rate(sr)
    audio_16k = audio.set_frame_rate(SR)
    audio_22k = np.array(audio_22k.get_array_of_samples(), dtype=np.float32)
    audio_16k = np.array(audio_16k.get_array_of_samples(), dtype=np.float32)
    return audio_22k, audio_16k


def main():
    
    st.set_page_config(
        page_title="speech-denoising-app",
        layout="wide"
    )
    
    logo_space, title_space, _, tooltip_space = st.columns([2.03, 5, 1, 0.75], gap="small")
    
    with logo_space:
        st.write(
            """
            <div style="display: flex; justify-content: left;">
                <b><span style="text-align: center; color: #101414; font-size: 10px"><br></span></b>
            </div>
            """,
            unsafe_allow_html=True
        )
        st.image('logo.png', width=48)
    
    with title_space:
        st.image('title.png', width=640)
        
    with tooltip_space:
        st.markdown(
            """
            <style>
            .tooltip {
                position: relative;
                display: inline-block;
                cursor: pointer;
                background-color: rgba(0, 76, 153, 1); /* Blue button color */
                padding: 10px;
                border-radius: 50%;
                font-size: 16px;
                font-weight: bold;
                width: 40px;
                height: 40px;
                text-align: center;
                line-height: 20px;
                color: white; /* Text color */
                box-shadow: 2px 2px 5px rgba(0, 0, 0, 0.2);
            }

            .tooltip .tooltiptext {
                visibility: hidden;
                width: 300px; /* Adjust width for readability */
                background-color: #333; /* Dark background for contrast */
                color: #fff;
                text-align: left; /* Align text to the left */
                border-radius: 8px;
                padding: 15px; /* Add padding for spacing */
                position: absolute;
                z-index: 1;
                top: 150%; /* Position below the button */
                left: 50%;
                transform: translateX(-50%);
                opacity: 0;
                transition: opacity 0.3s;
                font-size: 14px;
                line-height: 1.8; /* Adjust line height for readability */
                white-space: normal; /* Allow wrapping of text */
            }

            .tooltip:hover .tooltiptext {
                visibility: visible;
                opacity: 1;
            }
            </style>
            """,
            unsafe_allow_html=True,
        )

        st.markdown(
            """
            <div class="tooltip">
                ℹ
                <span class="tooltiptext">
                <strong>Steps:</strong><br>
                1) Denoise your own speech: Click <em>Start recording</em>, then <em>Stop recording</em> when you are finished.<br>
                2) Click <em>"Denoise"</em> and wait for a few seconds.<br>
                3) Both the original audio and denoised audio will be available for playback.<br><br>
                <strong>Note:</strong> Playing "noise" on your device while recording your speech to emulate speaking in a noisy environment will not work as intended. To do this emulation more realistically, play the noise on a different device (such as your phone) while recording your speech.
                </span>
            </div>
            """,
            unsafe_allow_html=True,
        )
        
    tab1, tab2 = st.tabs(["📂Denoise our samples speech", "🎙️Denoise your own speech"])
    
    with tab1:
        noisy_speech_files = load_noisy_speech()

        input_space_tab1, output_space_tab1 = st.columns([1, 1], gap="medium")
        _, _, _, compute_space_tab1= st.columns([0.7, 1, 1, 1], gap="small")
            
        with compute_space_tab1:
            compute_tab1 = st.button('Denoise', key='denoise_tab1')
        
        with input_space_tab1.container(height=CONTAINER_HEIGHT, border=True):
            lang_select_space, snr_select_space = st.columns([1, 1], gap="small")
            with lang_select_space:
                language_select = st.selectbox("Language", list(noisy_speech_files.keys()))
            with snr_select_space:
                if language_select:
                    snr_select = st.selectbox("SNR Level", list(noisy_speech_files[language_select].keys()))
            
            audio_path_tab1 = noisy_speech_files[language_select][snr_select]
            noisy_wav_22k_tab1, noisy_wav_tab1 = load_wav(audio_path_tab1)
            noisy_spec_tab1 = wav_to_spec(noisy_wav_22k_tab1, sr=22050)
            noisy_spec_buff_tab1 = export_spec_to_buffer(noisy_spec_tab1)
        
            st.audio(audio_path_tab1, format="wav")
            st.image(image=noisy_spec_buff_tab1)
            
        with output_space_tab1.container(height=CONTAINER_HEIGHT, border=True):
            st.write(
                """
                <div style="display: flex; justify-content: center;">
                    <b><span style="text-align: center; color: #808080; font-size: 51.5px">Output</span></b>
                </div>
                """,
                unsafe_allow_html=True
            )
            if noisy_wav_tab1.any() and compute_tab1:
                with st.spinner("Denoising..."):
                    denoised_wav_tab1 = denoise(noisy_wav_tab1)
                st.audio(denoised_wav_tab1, sample_rate=SR, format="audio/wav")
                denoised_spec_tab1 = wav_to_spec(denoised_wav_tab1, sr=SR)
                denoised_spec_buff_tab1 = export_spec_to_buffer(denoised_spec_tab1)
                st.image(image=denoised_spec_buff_tab1)
                
    with tab2:
        input_space_tab2, output_space_tab2 = st.columns([1, 1], gap="medium")
        _, record_space, _, compute_space_tab2 = st.columns([0.7, 1, 1, 1], gap="small")
        
        with record_space:
            record = recorder(
                start_prompt="Start Recording",
                stop_prompt="Stop Recording",
                just_once=False,
                use_container_width=False,
                format="wav",
                callback=None,
                args=(),
                kwargs={},
                key="tab2_recorder"
            )
            
        with compute_space_tab2:
            compute_tab2 = st.button('Denoise', key='denoise_tab2')
        
        noisy_wav_tab2 = np.array([])
        with input_space_tab2.container(height=CONTAINER_HEIGHT, border=True):
            st.write(
                """
                <div style="display: flex; justify-content: center;">
                    <b><span style="text-align: center; color: #808080; font-size: 51.5px">Input</span></b>
                </div>
                """,
                unsafe_allow_html=True
            )
            
            if record:
                wav_bytes_record = record['bytes']
                sr = record['sample_rate']
                noisy_wav_22k_tab2, noisy_wav_tab2 = process_recorded_wav_bytes(wav_bytes_record, sr=22050)
                noisy_spec_tab2 = wav_to_spec(noisy_wav_22k_tab2, sr=22050)
                noisy_spec_buff_tab2 = export_spec_to_buffer(noisy_spec_tab2)
                
                st.audio(wav_bytes_record, format="wav")
                st.image(image=noisy_spec_buff_tab2)
        
        with output_space_tab2.container(height=CONTAINER_HEIGHT, border=True):
            st.write(
                """
                <div style="display: flex; justify-content: center;">
                    <b><span style="text-align: center; color: #808080; font-size: 51.5px">Output</span></b>
                </div>
                """,
                unsafe_allow_html=True
            )
            if noisy_wav_tab2.any() and compute_tab2:
                with st.spinner("Denoising..."):
                    denoised_wav_tab2 = denoise(noisy_wav_tab2)
                st.audio(denoised_wav_tab2, sample_rate=SR, format="audio/wav")
                denoised_spec_tab2 = wav_to_spec(denoised_wav_tab2, sr=SR)
                denoised_spec_buff_tab2 = export_spec_to_buffer(denoised_spec_tab2)
                st.image(image=denoised_spec_buff_tab2)
                record = None
        

if __name__ == '__main__':
    main()