app.py

import os
from pathlib import Path
from httpx import AsyncClient

import gradio as gr
import numpy as np
from dotenv import load_dotenv
from fastrtc import (
    AdditionalOutputs,
    ReplyOnPause,
    Stream,
    audio_to_bytes,
    get_turn_credentials_async,
    get_turn_credentials,
)
from gradio.utils import get_space
from languages import LANGUAGES

cur_dir = Path(__file__).parent

load_dotenv()


client = AsyncClient(timeout=30)


async def transcribe_file(audio: tuple[int, np.ndarray], language: str):
    response = await client.post(
        url="https://cw18rfhfqf3db1m8.us-east-1.aws.endpoints.huggingface.cloud/api/v1/audio/transcriptions",
        headers={"Authorization": f"Bearer {os.getenv('HF_TOKEN')}"},
        files={"file": audio_to_bytes(audio)},
        data={"response_format": "text", "language": language},
    )
    return response.text


async def transcribe(audio: tuple[int, np.ndarray], transcript: str, language: str):
    text = await transcribe_file(audio, language)
    yield AdditionalOutputs(transcript + " " + text)


transcript = gr.Textbox(label="Transcript")
stream = Stream(
    ReplyOnPause(transcribe, input_sample_rate=48_100),
    modality="audio",
    mode="send",
    additional_inputs=[transcript, gr.Dropdown(choices=LANGUAGES, label="Language")],
    additional_outputs=[transcript],
    additional_outputs_handler=lambda a, b: b,
    rtc_configuration=get_turn_credentials_async,
    server_rtc_configuration=get_turn_credentials(ttl=604_800),
    concurrency_limit=20 if get_space() else None,
    time_limit=300, 
    ui_args={"title": ""},
)

iface = gr.Interface(
    fn=transcribe_file,
    inputs=[gr.Audio(label="Upload Audio", sources=["upload", "microphone"]), gr.Dropdown(choices=LANGUAGES, label="Language")],
    outputs=gr.Textbox(label="Transcript"),
)


with gr.Blocks() as demo:
    gr.HTML(
        """
        <h1 style='text-align: center; display: flex; align-items: center; justify-content: center;'>
        <img src="/gradio_api/file=AV_Huggy.png" alt="Streaming Huggy" style="height: 50px; margin-right: 10px"> Really Fast Whisper
        </h1>
    """
    )
    gr.HTML(
        """
    <h2 style='text-align: center'>
    Powered by <a href="https://huggingface.co/hfendpoints/whisper-large-v3">HF Inference Endpoints</a> and <a href="https://fastrtc.org/">FastRTC</a>
    </h2>
    """
    )
    with gr.Tabs():
        with gr.Tab("Streaming"):
            gr.Markdown(
                "Grant access to the microphone and speak naturally. The transcript will be updated as you pause."
            )
            stream.ui.render()
        with gr.Tab("File Upload"):
            iface.render()
if __name__ == "__main__":
    demo.launch(allowed_paths=["AV_Huggy.png"])