app.py

import gradio as gr
import os

import assemblyai as aai

import io
from scipy.io.wavfile import write


title = """<h1 align="center">🔥AssemblyAI: Conformer-2 Demo🔥</h1>"""

subtitle = (
    """<p align="center">Automatic Speech Recognition using the AssemblyAI API</p>"""
)
link = """<p align="center"><a href="https://www.assemblyai.com/blog/conformer-2/">Click here to learn more about the Conformer-2 model</a></p>"""


def submit_to_AAI(api_key, radio, audio_file, mic_recording):

    if radio == "Audio File":
        audio_data = audio_file
    elif radio == "Record Audio":
        audio_data = mic_recording

    if not api_key:
        return "Error! Did you use a valid API key?"

    aai.settings.api_key = api_key
    transcriber = aai.Transcriber()

    # Create temporary "file" and write data to it
    sr, aud = audio_data

    bytes_wav = bytes()
    temp_file = io.BytesIO(bytes_wav)
    write(temp_file, sr, aud)

    # Workaround to upload a file-like object before transcribing
    # This should be abstracted away in future SDK versions:
    try:
        upload_url = aai.api.upload_file(aai.Client.get_default().http_client, temp_file)
    except aai.types.TranscriptError as e:
        return str(e)

    # Now we can transcibe the url
    transcript = transcriber.transcribe(upload_url)

    if transcript.error is not None:
        return transcript.error

    paragraphs = transcript.get_paragraphs()
    return "\n\n".join(p.text for p in paragraphs)


def change_audio_source(radio):
    if radio == "Audio File":
        return [gr.Audio.update(visible=True), gr.Audio.update(visible=False)]
    elif radio == "Record Audio":
        return [gr.Audio.update(visible=False), gr.Audio.update(visible=True)]


with gr.Blocks(
    css="""#col_container {width: 1000px; margin-left: auto; margin-right: auto;}
                #chatbot {height: 520px; overflow: auto;}"""
) as demo:
    gr.HTML(
        '<center><a href="https://www.assemblyai.com/"><img src="file/images/logo.png" width="180px"></a></center>'
    )
    gr.HTML(title)
    gr.HTML(subtitle)
    gr.HTML(link)
    gr.HTML(
        """<center><a href="https://huggingface.co/spaces/assemblyai/Conformer1-Demo?duplicate=true"><img src="https://bit.ly/3gLdBN6" alt="Duplicate Space"></a>Duplicate the Space and run securely with your AssemblyAI API Key. <a href="https://www.assemblyai.com/dashboard/signup">Get a free key here.</a></center>"""
    )

    with gr.Column(elem_id="col_container"):
        api_key = gr.Textbox(
            type="password", label="Enter your AssemblyAI API key here"
        )

        with gr.Box():
            # Selector for audio source
            radio = gr.Radio(
                ["Audio File", "Record Audio"], label="Audio Source", value="Audio File"
            )
            # Audio object for both file and microphone data
            audio_file = gr.Audio()
            mic_recording = gr.Audio(source="microphone", visible=False)

            gr.Examples(
                [
                    os.path.join(os.path.dirname(__file__), "audio/audio_sample1.flac"),
                    os.path.join(
                        os.path.dirname(__file__), "audio/assemblyai_company.mp3"
                    ),
                ],
                audio_file,
            )

        btn = gr.Button("Run")

        out = gr.Textbox(
            placeholder="Your formatted transcript will appear here ...", lines=10
        )

        # Changing audio source changes Audio input component
        radio.change(
            fn=change_audio_source, inputs=[radio], outputs=[audio_file, mic_recording]
        )

        # Clicking "submit" uploads selected audio to AssemblyAI, performs requested analyses, and displays results
        btn.click(
            fn=submit_to_AAI,
            inputs=[api_key, radio, audio_file, mic_recording],
            outputs=out,
        )

    demo.queue(max_size=20, concurrency_count=10).launch(debug=True)