app.py

import os
from fastapi import FastAPI, File, Form, HTTPException
from fastapi.responses import JSONResponse
from pydantic import BaseModel
from groq import Groq
import io

# Set up the Groq client
os.environ["GROQ_API_KEY"] = os.getenv("GROQ_API_KEY")
client = Groq(api_key=os.environ["GROQ_API_KEY"])

app = FastAPI()

# Pydantic model for the transcription result
class TranscriptionResponse(BaseModel):
    transcription: str

class FeedbackResponse(BaseModel):
    grammar_feedback: str
    vocabulary_feedback: str
    grammar_score: int
    vocabulary_score: int

@app.get("/")
async def index():
    return {"message": "Welcome to the Audio Transcription API!"}

@app.post("/transcribe")
async def transcribe_audio(audio_data: bytes = File(...), language: str = Form(...)):
    try:
        # Transcribe the audio based on the selected language
        transcription = client.audio.transcriptions.create(
            file=("audio.wav", audio_data),
            model="whisper-large-v3",
            prompt="Transcribe the audio accurately based on the selected language.",
            response_format="text",
            language=language,
        )
        return JSONResponse(content=TranscriptionResponse(transcription=transcription).dict())
    except Exception as e:
        raise HTTPException(status_code=500, detail=str(e))

@app.post("/check_grammar")
async def check_grammar(transcription: str = Form(...), language: str = Form(...)):
    if not transcription or not language:
        raise HTTPException(status_code=400, detail="Missing transcription or language selection")

    try:
        # Grammar check
        grammar_prompt = (
            f"Briefly check the grammar of the following text in {language}: {transcription}. "
            "Identify any word that does not belong to the selected language and flag it. Based on the number of incorrect words, "
            "also check the grammar deeply and carefully. Provide a score from 1 to 10 based on the grammar accuracy, reducing points for incorrect words, "
            "and make sure to output the score on a new line after two line breaks like \"SCORE=\"."
        )

        grammar_check_response = client.chat.completions.create(
            model="llama3-groq-70b-8192-tool-use-preview",
            messages=[{"role": "user", "content": grammar_prompt}]
        )
        grammar_feedback = grammar_check_response.choices[0].message.content.strip()

        # Vocabulary check
        vocabulary_prompt = (
            f"Check the vocabulary accuracy of the following text in {language}: {transcription}. "
            "Identify any word that does not belong to the selected language and flag it. Based on the number of incorrect words, "
            "also check the grammar deeply and carefully. Provide a score from 1 to 10 based on the vocabulary accuracy, reducing points for incorrect words, "
            "and make sure to output the score on a new line after two line breaks like \"SCORE=\"."
        )

        vocabulary_check_response = client.chat.completions.create(
            model="llama-3.1-70b-versatile",
            messages=[{"role": "user", "content": vocabulary_prompt}]
        )
        vocabulary_feedback = vocabulary_check_response.choices[0].message.content.strip()

        # Calculate scores from feedback
        grammar_score = calculate_score(grammar_feedback)
        vocabulary_score = calculate_score(vocabulary_feedback)

        return JSONResponse(content=FeedbackResponse(
            grammar_feedback=grammar_feedback,
            vocabulary_feedback=vocabulary_feedback,
            grammar_score=grammar_score,
            vocabulary_score=vocabulary_score
        ).dict())
    except Exception as e:
        raise HTTPException(status_code=500, detail=str(e))

def calculate_score(feedback):
    """
    Calculate score based on feedback content. 
    This function searches for the keyword 'SCORE=' or similar variations 
    (SCORE:, score:, etc.) and extracts the score value.
    """
    import re
    match = re.search(r'(SCORE=|score=|SCORE:|score:|SCORE = )\s*(\d+)', feedback)
    
    if match:
        # Extract and return the score as an integer
        return int(match.group(2))
    
    # Return a default score of 0 if no score is found in the feedback
    return 0