app.py

from openai import OpenAI
from mistral_common.audio import Audio
from mistral_common.protocol.instruct.messages import AudioChunk, TextChunk, UserMessage
import gradio as gr
import httpx

# Patch transport to remove auth headers
class NoAuth(httpx.Auth):
    def auth_flow(self, request):
        yield request

client = OpenAI(
    api_key="dddd",
    base_url="http://localhost:8000/v1",
    http_client=httpx.Client(auth=NoAuth())
)

model = "mistralai/Voxtral-Mini-3B-2507"  # or just hardcode it

def transcribe(audio_file):
    if not audio_file:
        return "Please record something."

    with open(audio_file, "rb") as f:
        audio_bytes = f.read()
        
    audio = Audio.from_bytes(audio_bytes, strict=False)
    chunk = AudioChunk.from_audio(audio)
    prompt = TextChunk(text="Please transcribe this audio.")
    user_msg = UserMessage(content=[chunk, prompt]).to_openai()

    response = client.chat.completions.create(
        model=model,
        messages=[user_msg],
        temperature=0.0
    )

    return response.choices[0].message.content

gr.Interface(
    fn=transcribe,
    inputs=gr.Audio(type="filepath", label="Record your voice"),
    outputs=gr.Textbox(label="Transcription"),
    title="Transcribe Audio with Voxtral",
).launch()