app.py

# app.py
import os
import gradio as gr
from transformers import pipeline

hf_token = os.getenv("HUGGINGFACEHUB_API_TOKEN")  # this is None if not set

# 1. Speech-to-text (Whisper small)
asr = pipeline("automatic-speech-recognition", model="openai/whisper-small")

# 2. Text generation (LLM)
chatbot = pipeline("text-generation", model="mistralai/Mistral-7B-Instruct-v0.2", use_auth_token=hf_token)

# 3. Text-to-speech (Parler-TTS mini)
tts = pipeline("text-to-speech", model="parler-tts/parler-tts-mini-multilingual")

def voice_interview(audio):
    # audio is a (sample_rate, numpy array)
    transcript = asr(audio)["text"]
    reply = chatbot(transcript, max_length=200, do_sample=True)[0]["generated_text"]
    audio_out = tts(reply)  # returns dict with 'audio' key
    return transcript, reply, (audio_out["audio"], audio_out["sampling_rate"])

with gr.Blocks() as demo:
    gr.Markdown("### 🎤 Mock Interview Bot")
    audio_in = gr.Audio(sources=["microphone"], type="numpy", label="Speak your answer")
    transcript_out = gr.Textbox(label="Transcript")
    reply_out = gr.Textbox(label="Bot Reply")
    audio_out = gr.Audio(label="Bot's Voice Reply")
    btn = gr.Button("Submit")

    btn.click(fn=voice_interview, inputs=audio_in, outputs=[transcript_out, reply_out, audio_out])

demo.launch()