app.py

import gradio as gr
import whisper
from datetime import datetime
from PIL import Image
import flag
import os
import pandas as pd
import matplotlib.pyplot as plt
import numpy as np
from sklearn.preprocessing import LabelEncoder
import seaborn as sns

plt.switch_backend('Agg')
pd.options.display.max_columns = 25
pd.options.display.max_rows = 300

stable_diffusion = gr.Blocks.load(name="spaces/stabilityai/stable-diffusion")
code_generation=gr.Blocks.load(name="spaces/THUDM/CodeGeeX")
### ————————————————————————————————————————

title="DataTeller"

### ————————————————————————————————————————

whisper_model = whisper.load_model("small")


def get_images(prompt):
    gallery_dir = stable_diffusion(prompt, fn_index=2)
    return [os.path.join(gallery_dir, img) for img in os.listdir(gallery_dir)]

def get_code(prompt):
    examples_dir = code_generation(prompt, fn_index=0)
    return [os.path.join(examples_dir, output) for output in os.listdir(examples_dir)]



def magic_whisper_to_sd(audio, guidance_scale, nb_iterations, seed):

    whisper_results = translate(audio)
    prompt = whisper_results[2]
    images = get_images(prompt)

    return whisper_results[0], whisper_results[1], whisper_results[2], images


def generate_code(audio):

    whisper_results = translate(audio)
    prompt = whisper_results[2]
    code = get_code(prompt)

    return code

def translate(audio):
    print("""
    —
    Sending audio to Whisper ...
    —
    """)
    now = datetime.now()
    date_time_str = now.strftime("%Y-%m-%d %H:%M:%S")
    print('DateTime String:', date_time_str)

    audio = whisper.load_audio(audio)
    audio = whisper.pad_or_trim(audio)

    mel = whisper.log_mel_spectrogram(audio).to(whisper_model.device)

    _, probs = whisper_model.detect_language(mel)

    transcript_options = whisper.DecodingOptions(task="transcribe", fp16 = False)
    translate_options = whisper.DecodingOptions(task="translate", fp16 = False)

    transcription = whisper.decode(whisper_model, mel, transcript_options)
    translation = whisper.decode(whisper_model, mel, translate_options)

    print("language spoken: " + transcription.language)
    print("transcript: " + transcription.text)
    print("———————————————————————————————————————————")
    print("translated: " + translation.text)
    if transcription.language == "en":
        tr_flag = flag.flag('GB')
    else:
        tr_flag = flag.flag(transcription.language)
    return tr_flag, transcription.text, translation.text


### ————————————————————————————————————————

css = """
        .container {
            max-width: 880px;
            margin: auto;
            padding-top: 1.5rem;
        }
        a {
            text-decoration: underline;
        }
        h1 {
            font-weight: 900;
            margin-bottom: 7px;
            text-align: center;
            font-size: 2em;
            margin-bottom: 1em;
        }
        #w2sd_container{
            margin-top: 20px;
        }
        .footer {
            margin-bottom: 45px;
            margin-top: 35px;
            text-align: center;
            border-bottom: 1px solid #e5e5e5;
        }
        .footer>p {
            font-size: .8rem;
            display: inline-block;
            padding: 0 10px;
            transform: translateY(10px);
            background: white;
        }
        .dark .footer {
            border-color: #303030;
        }
        .dark .footer>p {
            background: #0b0f19;
        }
        .tabitem {
            border-bottom-left-radius: 10px;
            border-bottom-right-radius: 10px;
        }
        #record_tab, #upload_tab {
            font-size: 1.2em;
        }
        #record_btn{

        }
        #record_btn > div > button > span {
            width: 2.375rem;
            height: 2.375rem;
        }
        #record_btn > div > button > span > span {
            width: 2.375rem;
            height: 2.375rem;
        }
        audio {
            margin-bottom: 10px;
        }
        div#record_btn > .mt-6{
            margin-top: 0!important;
        }
        div#record_btn > .mt-6 button {
            font-size: 2em;
            width: 100%;
            padding: 20px;
            height: 160px;
        }
        div#upload_area {
            height: 11.1rem;
        }
        div#upload_area > div.w-full > div {
            min-height: 9rem;
        }
        #check_btn_1, #check_btn_2{
            color: #fff;
            --tw-gradient-from: #4caf50;
            --tw-gradient-stops: var(--tw-gradient-from), var(--tw-gradient-to);
            --tw-gradient-to: #4caf50;
            border-color: #8bc34a;
        }
        #magic_btn_1, #magic_btn_2{
            color: #fff;
            --tw-gradient-from: #f44336;
            --tw-gradient-stops: var(--tw-gradient-from), var(--tw-gradient-to);
            --tw-gradient-to: #ff9800;
            border-color: #ff9800;
        }
        input::-webkit-inner-spin-button, input::-webkit-outer-spin-button {
            -webkit-appearance: none;
        }
        input[type=number] {
            -moz-appearance: textfield;
        }
        input[type=range] {
            -webkit-appearance: none;
            cursor: pointer;
            height: 1px;
            background: currentColor;
        }
        input[type=range]::-webkit-slider-thumb {
            -webkit-appearance: none;
            width: 0.5em;
            height: 1.2em;
            border-radius: 10px;
            background: currentColor;
        }
        input[type=range]::-moz-range-thumb{
            width: 0.5em;
            height: 1.2em;
            border-radius: 10px;
            background: currentColor;
        }
        div#spoken_lang textarea {
            font-size: 4em;
            line-height: 1em;
            text-align: center;
        }
        div#transcripted {
            flex: 4;
        }
        div#translated textarea {
            font-size: 1.5em;
            line-height: 1.25em;
        }
        #sd_settings {
            margin-bottom: 20px;
        }
        #diffuse_btn {
            color: #fff;
            font-size: 1em;
            margin-bottom: 20px;
            --tw-gradient-from: #4caf50;
            --tw-gradient-stops: var(--tw-gradient-from), var(--tw-gradient-to);
            --tw-gradient-to: #4caf50;
            border-color: #8bc34a;
        }
         #codegen_btn {
            color: #fff;
            font-size: 1em;
            margin-bottom: 20px;
            --tw-gradient-from: #4caf50;
            --tw-gradient-stops: var(--tw-gradient-from), var(--tw-gradient-to);
            --tw-gradient-to: #4caf50;
            border-color: #8bc34a;
        }
        #notice {
            padding: 20px 14px 10px;
            display: flex;
            align-content: space-evenly;
            gap: 20px;
            line-height: 1em;
            font-size: .8em;
            border: 1px solid #374151;
            border-radius: 10px;
        }
        #about {
            padding: 20px;
        }
        #notice > div {
            flex: 1;
        }

"""

### ————————————————————————————————————————

with gr.Blocks(css=css) as demo:
    with gr.Column():
        gr.HTML('''
            <h1>
                DataTeller
            </h1>
            <p style='text-align: center;'>
                Generate data visualizations by speaking in your native language ! Try it in Romanian 😉
            </p>

            <p style='text-align: center;'>
                This demo is wired to the official SD Space and using the Whisperer model
            </p>

        ''')


        gr.Markdown(
            """

            ## 1. Record audio or Upload an audio file:
            """
        )

        with gr.Tab(label="Record audio input", elem_id="record_tab"):
            with gr.Column():
                record_input = gr.Audio(
                                    source="microphone",
                                    type="filepath",
                                    show_label=False,
                                    elem_id="record_btn"
                                )
                with gr.Row():
                    audio_r_translate = gr.Button("Check Whisper first ? 👍", elem_id="check_btn_1")
                    audio_r_direct_sd = gr.Button("Magic Whisper › SD right now!", elem_id="magic_btn_1")

        with gr.Tab(label="Upload audio input", elem_id="upload_tab"):
            with gr.Column():
                upload_input = gr.Audio(
                                    source="upload",
                                    type="filepath",
                                    show_label=False,
                                    elem_id="upload_area"
                                )
                with gr.Row():
                    audio_u_translate = gr.Button("Check Whisper first ? 👍", elem_id="check_btn_2")
                    # audio_u_direct_sd = gr.Button("Magic Whisper › SD right now!", elem_id="magic_btn_2")

        with gr.Accordion(label="Stable Diffusion Settings", elem_id="sd_settings", visible=False):
            with gr.Row():
                guidance_scale = gr.Slider(2, 15, value = 7, label = 'Guidance Scale')
                nb_iterations = gr.Slider(10, 50, value = 25, step = 1, label = 'Steps')
                seed = gr.Slider(label = "Seed", minimum = 0, maximum = 2147483647, step = 1, randomize = True)

        gr.Markdown(
            """
            ## 2. Check Whisper output, correct it if necessary:
            """
        )

        with gr.Row():

            transcripted_output = gr.Textbox(
                                    label="Transcription in your detected spoken language",
                                    lines=3,
                                    elem_id="transcripted"
                                )
            language_detected_output = gr.Textbox(label="Native language", elem_id="spoken_lang",lines=3)

        with gr.Column():
            translated_output = gr.Textbox(
                                    label="Transcript translated in English by Whisper",
                                    lines=4,
                                    elem_id="translated"
                                )
            with gr.Row():
                clear_btn = gr.Button(value="Clear")
                diffuse_btn = gr.Button(value="OK, Diffuse this prompt !", elem_id="diffuse_btn")
                codegen_btn = gr.Button(value="Generate code from this prompt!", elem_id="codegen_btn")

                clear_btn.click(fn=lambda value: gr.update(value=""), inputs=clear_btn, outputs=translated_output)

        gr.Markdown(
            """
            ## 3. Use whisper results for code generation
            """
        )
        with gr.Column():

            with gr.Row():
                # clear_btn = gr.Button(value="Clear")
                # diffuse_btn = gr.Button(value="Diffuse this prompt!", elem_id="diffuse_btn")
                # codegen_btn = gr.Button(value="Generate code from this prompt!", elem_id="codegen_btn")

                code = gr.Textbox(
                                    label="Code generated",
                                    lines=4,
                                    elem_id="translated"
                                )
                # clear_btn.click(fn=lambda value: gr.update(value=""), inputs=clear_btn, outputs=translated_output)

        # gr.Markdown("""
        #     ## 4. Plot the generated code  ☕️
        #     Inference time is about ~20-30 seconds, when it's your turn 😬
        #     """
        #     )
        # with gr.Column():
        #     with gr.Row():
        #         def outbreak(plot_type):
        #             df = pd.read_csv('emp_experience_data.csv')
        #             data_encoded = df.copy(deep=True)
        #             categorical_column = ['Attrition', 'Gender', 'BusinessTravel', 'Education', 'EmployeeExperience', 'EmployeeFeedbackSentiments', 'Designation',
        #                                     'SalarySatisfaction', 'HealthBenefitsSatisfaction', 'UHGDiscountProgramUsage', 'HealthConscious', 'CareerPathSatisfaction', 'Region']
        #             label_encoding = LabelEncoder()
        #             for col in categorical_column:
        #                 data_encoded[col] = label_encoding.fit_transform(data_encoded[col])

        #             if plot_type == "Find Data Correlation":
        #                 fig = plt.figure()
        #                 data_correlation = data_encoded.corr()
        #                 sns.heatmap(data_correlation, xticklabels = data_correlation.columns, yticklabels = data_correlation.columns)
        #                 return fig
        #             if plot_type == "Filter Correlation Data":
        #                 fig = plt.figure()
        #                 filtered_df = df[['EmployeeExperience', 'EmployeeFeedbackSentiments', 'Age', 'SalarySatisfaction', 'BusinessTravel', 'HealthBenefitsSatisfaction']]
        #                 correlation_filter_data = filtered_df.corr()
        #                 sns.heatmap(correlation_filter_data, xticklabels = filtered_df.columns, yticklabels = filtered_df.columns)
        #                 return fig
        #             if plot_type == "Age vs Attrition":
        #                 fig = plt.figure()
        #                 plt.hist(data_encoded['Age'], bins=np.arange(0,80,10), alpha=0.8, rwidth=0.9, color='red')
        #                 plt.xlabel("Age")
        #                 plt.ylabel("Count")
        #                 plt.title("Age vs Attrition")
        #                 return fig
        #             if plot_type == "Business Travel vs Attrition":
        #                 fig = plt.figure()
        #                 ax = sns.countplot(x="BusinessTravel", hue="Attrition", data=data_encoded)
        #                 for p in ax.patches:
        #                     ax.annotate('{}'.format(p.get_height()), (p.get_x(), p.get_height()+1))
        #                 return fig
        #             if plot_type == "Employee Experience vs Attrition":
        #                 fig = plt.figure()
        #                 ax = sns.countplot(x="EmployeeExperience", hue="Attrition", data=data_encoded)
        #                 for p in ax.patches:
        #                     ax.annotate('{}'.format(p.get_height()), (p.get_x(), p.get_height()+1))
        #                 return figure

    #         inputs = [
    #             gr.Dropdown(["Find Data Correlation", "Filter Correlation Data", "Business Travel vs Attrition", "Employee Experience vs Attrition", "Age vs Attrition",], label="Data Correlation and Visualization")
    #         ]
    #         outputs = gr.Plot()

    #         demo2 = gr.Interface(
    #             fn = outbreak,
    #             inputs = inputs,
    #             outputs = outputs,
    #             title="Employee-Experience: Data Correlation and Pattern Visualization",
    #             allow_flagging=False
    #         )


        gr.Markdown("""
            ## 5. Wait for Stable Diffusion Results ☕️
            Inference time is about ~20-30 seconds, when it's your turn 😬
            """
            )

        sd_output = gr.Gallery().style(grid=2, height="auto")


        gr.Markdown("""
            ### 📌 About the models
            <p style='font-size: 1em;line-height: 1.5em;'>
            <strong>Whisper</strong> is a general-purpose speech recognition model.<br /><br />
            It is trained on a large dataset of diverse audio and is also a multi-task model that can perform multilingual speech recognition as well as speech translation and language identification. <br />
            —
            </p>
            <p style='font-size: 1em;line-height: 1.5em;'>
            <strong>Stable Diffusion</strong> is a state of the art text-to-image model that generates images from text.
            </p>
            <div id="notice">
                <div>
                LICENSE
                <p style='font-size: 0.8em;'>
                The model is licensed with a <a href="https://huggingface.co/spaces/CompVis/stable-diffusion-license" target="_blank">CreativeML Open RAIL-M</a> license.</p>
                <p style='font-size: 0.8em;'>
                The authors claim no rights on the outputs you generate, you are free to use them and are accountable for their use which must not go against the provisions set in this license.</p>
                <p style='font-size: 0.8em;'>
                The license forbids you from sharing any content that violates any laws, produce any harm to a person, disseminate any personal information that would be meant for harm, spread misinformation and target vulnerable groups.</p>
                <p style='font-size: 0.8em;'>
                For the full list of restrictions please <a href="https://huggingface.co/spaces/CompVis/stable-diffusion-license" target="_blank" target="_blank">read the license</a>.
                 </p>
                 </div>
                 <div>
                 Biases and content acknowledgment
                 <p style='font-size: 0.8em;'>
                 Despite how impressive being able to turn text into image is, beware to the fact that this model may output content that reinforces or exacerbates societal biases, as well as realistic faces, pornography and violence.</p>
                <p style='font-size: 0.8em;'>
                The model was trained on the <a href="https://laion.ai/blog/laion-5b/" target="_blank">LAION-5B dataset</a>, which scraped non-curated image-text-pairs from the internet (the exception being the removal of illegal content) and is meant for research purposes.</p>
                <p style='font-size: 0.8em;'>  You can read more in the <a href="https://huggingface.co/CompVis/stable-diffusion-v1-4" target="_blank">model card</a>.
                 </p>
                 </div>
             </div>

        """, elem_id="about")

        audio_r_translate.click(translate,
                                inputs = record_input,
                                outputs = [
                                    language_detected_output,
                                    transcripted_output,
                                    translated_output
                                ])

        audio_u_translate.click(translate,
                                inputs = upload_input,
                                outputs = [
                                    language_detected_output,
                                    transcripted_output,
                                    translated_output
                                ])

        audio_r_direct_sd.click(magic_whisper_to_sd,
                                inputs = [
                                    record_input,
                                    guidance_scale,
                                    nb_iterations,
                                    seed
                                ],
                                outputs = [
                                    language_detected_output,
                                    transcripted_output,
                                    translated_output,
                                    sd_output
                                ])

        # audio_u_direct_sd.click(magic_whisper_to_sd,
        #                         inputs = [
        #                             upload_input,
        #                             guidance_scale,
        #                             nb_iterations,
        #                             seed
        #                         ],
        #                         outputs = [
        #                             language_detected_output,
        #                             transcripted_output,
        #                             translated_output,
        #                             sd_output
        #                         ])

        diffuse_btn.click(get_images,
                              inputs = [
                                  translated_output
                                  ],
                              outputs = sd_output
                          )
        codegen_btn.click(generate_code,
                              inputs = [
                                  translated_output
                                  ],
                              outputs = code
                          )
        gr.HTML('''
                <div class="footer">
                    <p>Whisper by <a href="https://github.com/openai/whisper" target="_blank">OpenAI</a> - Stable Diffusion by <a href="https://huggingface.co/CompVis" target="_blank">CompVis</a> and <a href="https://huggingface.co/stabilityai"  target="_blank">Stability AI</a>
                    </p>
                </div>
                ''')


if __name__ == "__main__":
    demo.launch()