Add speaking endpoints and clean code.

.env

@@ -1,3 +1,4 @@
 OPENAI_API_KEY=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
 JWT_SECRET_KEY=6e9c124ba92e8814719dcb0f21200c8aa4d0f119a994ac5e06eb90a366c83ab2
 JWT_TEST_TOKEN=eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyJzdWIiOiJ0ZXN0In0.Emrs2D3BmMP4b3zMjw0fJTPeyMwWEBDbxx2vvaWguO0
+GOOGLE_APPLICATION_CREDENTIALS=firebase-configs/mti-ielts-626a2dcf6091.json

app.py

@@ -1,9 +1,15 @@
 from flask import Flask, request
 from flask_jwt_extended import JWTManager, jwt_required
 from functools import reduce
+import firebase_admin
+from firebase_admin import credentials
+from helper.api_messages import QuestionType, get_grading_messages, get_question_gen_messages
+from helper.firebase_helper import download_firebase_file
+from helper.speech_to_text_helper import speech_to_text
 from helper.token_counter import count_tokens
 from helper.openai_interface import make_openai_call
 import os
+import uuid
 
 from dotenv import load_dotenv
 
@@ -14,91 +20,80 @@ app = Flask(__name__)
 app.config['JWT_SECRET_KEY'] = os.getenv("JWT_SECRET_KEY")
 jwt = JWTManager(app)
 
+# Initialize Firebase Admin SDK
+cred = credentials.Certificate(os.getenv("GOOGLE_APPLICATION_CREDENTIALS"))
+firebase_admin.initialize_app(cred)
+
 GRADING_TEMPERATURE = 0.1
 GEN_QUESTION_TEMPERATURE = 0.7
 WRITING_TASK_2_POST_FIELDS = ['overall', 'comment', 'task_response']
 WRITING_TASK_2_GET_FIELDS = ['question']
 
+FIREBASE_BUCKET = 'mti-ielts.appspot.com'
+AUDIO_FILES_PATH = 'download-audio/'
+
+
 @app.route('/writing_task2', methods=['POST'])
 @jwt_required()
 def grade_writing_task():
-    data = request.get_json()  # Assuming the request data is in JSON format
+    try:
-    question = data.get('question')
+        data = request.get_json()
-    answer = data.get('answer')
+        question = data.get('question')
-    messages = [
+        answer = data.get('answer')
-        {
+        messages = get_grading_messages(QuestionType.WRITING_TASK_2, question, answer)
-            "role": "system",
+        token_count = reduce(lambda count, item: count + count_tokens(item)['n_tokens'],
-            "content": "You are a IELTS examiner.",
+                             map(lambda x: x["content"], filter(lambda x: "content" in x, messages)), 0)
-        },
+        response = make_openai_call(messages, token_count, WRITING_TASK_2_POST_FIELDS, GRADING_TEMPERATURE)
-        {
+        return response
-            "role": "system",
+    except Exception as e:
-            "content": f"The question you have to grade is of type Writing Task 2 and is the following: {question}",
+        return str(e)
-        },
+
-        {
-            "role": "user",
-            "content": "It is mandatory for you to provide your response with the overall grade and breakdown grades, "
-                       "in the following json format: {'comment': 'comment about answer quality', 'overall': 7.0, 'task_response': {'Task Achievement': 8.0, "
-                       "'Coherence and Cohesion': 6.5, 'Lexical Resource': 7.5, 'Grammatical Range and Accuracy': "
-                       "6.0}}",
-        },
-        {
-            "role": "user",
-            "content": "Example output: { 'comment': 'Overall, the response is good but there are some areas that need "
-                       "improvement.\n\nIn terms of Task Achievement, the writer has addressed all parts of the question "
-                       "and has provided a clear opinion on the topic. However, some of the points made are not fully "
-                       "developed or supported with examples.\n\nIn terms of Coherence and Cohesion, there is a clear "
-                       "structure to the response with an introduction, body paragraphs and conclusion. However, there "
-                       "are some issues with cohesion as some sentences do not flow smoothly from one to another.\n\nIn "
-                       "terms of Lexical Resource, there is a good range of vocabulary used throughout the response and "
-                       "some less common words have been used effectively.\n\nIn terms of Grammatical Range and Accuracy, "
-                       "there are some errors in grammar and sentence structure which affect clarity in places.\n\nOverall, "
-                       "this response would score a band 6.5.', 'overall': 6.5, 'task_response': "
-                       "{ 'Coherence and Cohesion': 6.5, 'Grammatical Range and Accuracy': 6.0, 'Lexical Resource': 7.0, "
-                       "'Task Achievement': 7.0}}",
-        },
-        {
-            "role": "user",
-            "content": f"Evaluate this answer according to ielts grading system: {answer}",
-        },
-    ]
-    token_count = reduce(lambda count, item: count + count_tokens(item)['n_tokens'],
-                         map(lambda x: x["content"], filter(lambda x: "content" in x, messages)), 0)
-    response = make_openai_call(messages, token_count, WRITING_TASK_2_POST_FIELDS, GRADING_TEMPERATURE)
-    return response
 
 @app.route('/writing_task2', methods=['GET'])
 @jwt_required()
 def get_writing_task_question():
-    messages = [
+    try:
-        {
+        messages = get_question_gen_messages(QuestionType.WRITING_TASK_2)
-            "role": "system",
+        token_count = reduce(lambda count, item: count + count_tokens(item)['n_tokens'],
-            "content": "You are a IELTS program that generates questions for the exams.",
+                             map(lambda x: x["content"], filter(lambda x: "content" in x, messages)), 0)
-        },
+        response = make_openai_call(messages, token_count, WRITING_TASK_2_GET_FIELDS, GEN_QUESTION_TEMPERATURE)
-        {
+        return response
-            "role": "system",
+    except Exception as e:
-            "content": "The question you have to generate is of type Writing Task 2 and is the following.",
+        return str(e)
-        },
+
-        {
+@app.route('/speaking_task', methods=['POST'])
-            "role": "user",
+@jwt_required()
-            "content": "It is mandatory for you to provide your response with the question "
+def grade_speaking_task():
-                       "in the following json format: {'question': 'question'}",
+    sound_file_name = AUDIO_FILES_PATH + str(uuid.uuid4())
-        },
+    try:
-        {
+        data = request.get_json()
-            "role": "user",
+        question = data.get('question')
-            "content": "Example output: { 'question': 'We are becoming increasingly dependent on computers. "
+        answer_firebase_path = data.get('answer')
-                       "They are used in businesses, hospitals, crime detection and even to fly planes. What things will "
+
-                       "they be used for in the future? Is this dependence on computers a good thing or should we he more "
+        download_firebase_file(FIREBASE_BUCKET, answer_firebase_path, sound_file_name)
-                       "auspicious of their benefits?'}",
+        answer = speech_to_text(sound_file_name)
-        },
+
-        {
+        messages = get_grading_messages(QuestionType.SPEAKING, question, answer)
-            "role": "user",
+        token_count = reduce(lambda count, item: count + count_tokens(item)['n_tokens'],
-            "content": "Generate a question for IELTS exam Writing Task 2.",
+                             map(lambda x: x["content"], filter(lambda x: "content" in x, messages)), 0)
-        },
+        response = make_openai_call(messages, token_count, WRITING_TASK_2_POST_FIELDS, GRADING_TEMPERATURE)
-    ]
+        os.remove(sound_file_name)
-    token_count = reduce(lambda count, item: count + count_tokens(item)['n_tokens'],
+        return response
-                         map(lambda x: x["content"], filter(lambda x: "content" in x, messages)), 0)
+    except Exception as e:
-    response = make_openai_call(messages, token_count, WRITING_TASK_2_GET_FIELDS, GEN_QUESTION_TEMPERATURE)
+        os.remove(sound_file_name)
-    return response
+        return str(e)
+
+@app.route('/speaking_task', methods=['GET'])
+@jwt_required()
+def get_speaking_task_question():
+    try:
+        messages = get_question_gen_messages(QuestionType.SPEAKING)
+        token_count = reduce(lambda count, item: count + count_tokens(item)['n_tokens'],
+                             map(lambda x: x["content"], filter(lambda x: "content" in x, messages)), 0)
+        response = make_openai_call(messages, token_count, WRITING_TASK_2_GET_FIELDS, GEN_QUESTION_TEMPERATURE)
+        return response
+    except Exception as e:
+        return str(e)
 
 
 if __name__ == '__main__':

firebase-configs/mti-ielts-626a2dcf6091.json

@@ -0,0 +1,13 @@
+{
+  "type": "service_account",
+  "project_id": "mti-ielts",
+  "private_key_id": "626a2dcf60916a1b5011f388495b8f9c4fc065ef",
+  "private_key": "-----BEGIN PRIVATE KEY-----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\nTQqp88pfTbOnkdJBVi0AWs5B\n-----END PRIVATE KEY-----\n",
+  "client_email": "firebase-adminsdk-dyg6p@mti-ielts.iam.gserviceaccount.com",
+  "client_id": "104980563453519094431",
+  "auth_uri": "https://accounts.google.com/o/oauth2/auth",
+  "token_uri": "https://oauth2.googleapis.com/token",
+  "auth_provider_x509_cert_url": "https://www.googleapis.com/oauth2/v1/certs",
+  "client_x509_cert_url": "https://www.googleapis.com/robot/v1/metadata/x509/firebase-adminsdk-dyg6p%40mti-ielts.iam.gserviceaccount.com",
+  "universe_domain": "googleapis.com"
+}

helper/api_messages.py

@@ -0,0 +1,143 @@
+from enum import Enum
+
+
+class QuestionType(Enum):
+    WRITING_TASK_2 = "Writing Task 2"
+    SPEAKING = "Speaking Task"
+
+
+def get_grading_messages(question_type: QuestionType, question: str, answer: str):
+    if QuestionType.WRITING_TASK_2 == question_type:
+        return [
+            {
+                "role": "system",
+                "content": "You are a IELTS examiner.",
+            },
+            {
+                "role": "system",
+                "content": f"The question you have to grade is of type Writing Task 2 and is the following: {question}",
+            },
+            {
+                "role": "user",
+                "content": "It is mandatory for you to provide your response with the overall grade and breakdown grades, "
+                           "with just the following json format: {'comment': 'comment about answer quality', 'overall': 7.0, "
+                           "'task_response': {'Task Achievement': 8.0, 'Coherence and Cohesion': 6.5, 'Lexical Resource': 7.5, "
+                           "'Grammatical Range and Accuracy': 6.0}}",
+            },
+            {
+                "role": "user",
+                "content": "Example output: { 'comment': 'Overall, the response is good but there are some areas that need "
+                           "improvement.\n\nIn terms of Task Achievement, the writer has addressed all parts of the question "
+                           "and has provided a clear opinion on the topic. However, some of the points made are not fully "
+                           "developed or supported with examples.\n\nIn terms of Coherence and Cohesion, there is a clear "
+                           "structure to the response with an introduction, body paragraphs and conclusion. However, there "
+                           "are some issues with cohesion as some sentences do not flow smoothly from one to another.\n\nIn "
+                           "terms of Lexical Resource, there is a good range of vocabulary used throughout the response and "
+                           "some less common words have been used effectively.\n\nIn terms of Grammatical Range and Accuracy, "
+                           "there are some errors in grammar and sentence structure which affect clarity in places.\n\nOverall, "
+                           "this response would score a band 6.5.', 'overall': 6.5, 'task_response': "
+                           "{ 'Coherence and Cohesion': 6.5, 'Grammatical Range and Accuracy': 6.0, 'Lexical Resource': 7.0, "
+                           "'Task Achievement': 7.0}}",
+            },
+            {
+                "role": "user",
+                "content": f"Evaluate this answer according to ielts grading system: {answer}",
+            },
+        ]
+    elif QuestionType.SPEAKING == question_type:
+        return [
+            {
+                "role": "user",
+                "content": "You are a IELTS examiner.",
+            },
+            {
+                "role": "user",
+                "content": f"The question you have to grade is of type Speaking and is the following: {question}",
+            },
+            {
+                "role": "user",
+                "content": "It is mandatory for you to provide your response with the overall grade and breakdown grades, "
+                           "with just the following json format: {'comment': 'comment about answer quality', 'overall': 7.0, "
+                           "'task_response': {'Fluency and Coherence': 8.0, 'Lexical Resource': 6.5, "
+                           "'Grammatical Range and Accuracy': 7.5, 'Pronunciation': 6.0}}",
+            },
+            {
+                "role": "user",
+                "content": "Example output: { 'comment': 'The candidate has provided a clear response to the question and has "
+                           "given examples of how they spend their weekends. However, there are some issues with grammar and "
+                           "pronunciation that affect the overall score. In terms of fluency and coherence, the candidate speaks "
+                           "clearly and smoothly with only minor hesitations. They have also provided a well-organized response "
+                           "that is easy to follow. Regarding lexical resource, the candidate has used a range of vocabulary "
+                           "related to weekend activities but there are some errors in word choice that affect the meaning of "
+                           "their sentences. In terms of grammatical range and accuracy, the candidate has used a mix of simple "
+                           "and complex sentence structures but there are some errors in subject-verb agreement and preposition "
+                           "use. Finally, regarding pronunciation, the candidate's speech is generally clear but there are some "
+                           "issues with stress and intonation that make it difficult to understand at times.', 'overall': 6.5, "
+                           "'task_response': {'Fluency and Coherence': 7.0, 'Lexical Resource': 6.5, 'Grammatical Range and Accuracy': 7.0,"
+                           " 'Pronunciation': 6.0}}",
+            },
+            {
+                "role": "user",
+                "content": f"Evaluate this answer according to ielts grading system: {answer}",
+            },
+        ]
+    else:
+        raise Exception("Question type not implemented: " + question_type.value)
+
+
+def get_question_gen_messages(question_type: QuestionType):
+    if QuestionType.WRITING_TASK_2 == question_type:
+        return [
+            {
+                "role": "system",
+                "content": "You are a IELTS program that generates questions for the exams.",
+            },
+            {
+                "role": "system",
+                "content": "The question you have to generate is of type Writing Task 2.",
+            },
+            {
+                "role": "user",
+                "content": "It is mandatory for you to provide your response with the question "
+                           "just with the following json format: {'question': 'question'}",
+            },
+            {
+                "role": "user",
+                "content": "Example output: { 'question': 'We are becoming increasingly dependent on computers. "
+                           "They are used in businesses, hospitals, crime detection and even to fly planes. What things will "
+                           "they be used for in the future? Is this dependence on computers a good thing or should we he more "
+                           "auspicious of their benefits?'}",
+            },
+            {
+                "role": "user",
+                "content": "Generate a question for IELTS exam Writing Task 2.",
+            },
+        ]
+    elif QuestionType.SPEAKING == question_type:
+        return [
+            {
+                "role": "system",
+                "content": "You are a IELTS program that generates questions for the exams.",
+            },
+            {
+                "role": "system",
+                "content": "The question you have to generate is of type Speaking Task.",
+            },
+            {
+                "role": "user",
+                "content": "It is mandatory for you to provide your response with the question "
+                           "just with the following json format: {'question': 'question'}",
+            },
+            {
+                "role": "user",
+                "content": "Example output: { 'question': 'Describe someone you know who does something well. You should say "
+                           "who this person is, how do you know this person, what they do well and explain why you think this "
+                           "person is so good at doing this.'}",
+            },
+            {
+                "role": "user",
+                "content": "Generate a question for IELTS exam Speaking Task.",
+            },
+        ]
+    else:
+        raise Exception("Question type not implemented: " + question_type.value)

helper/firebase_helper.py

@@ -0,0 +1,15 @@
+from google.cloud import storage
+
+def download_firebase_file(bucket_name, source_blob_name, destination_file_name):
+    # Downloads a file from Firebase Storage.
+    storage_client = storage.Client()
+    bucket = storage_client.bucket(bucket_name)
+    blob = bucket.blob(source_blob_name)
+    blob.download_to_filename(destination_file_name)
+    print(f"File downloaded to {destination_file_name}")
+    return destination_file_name
+
+
+
+
+

helper/openai_interface.py

@@ -1,6 +1,7 @@
 import json
 import openai
 import os
+import re
 
 from dotenv import load_dotenv
 
@@ -16,15 +17,25 @@ TRY_LIMIT = 1
 
 try_count = 0
 def process_response(input_string):
-    json_obj = {}
+    if '{' in input_string:
-    parsed_string = input_string.replace("'", "\"")
+        try:
-    parsed_string = parsed_string.replace("\n\n", " ")
+            # Find the index of the first occurrence of '{'
-    try:
+            index = input_string.index('{')
-        json_obj = json.loads(parsed_string)
+            # Extract everything after the first '{' (inclusive)
-    except json.JSONDecodeError:
+            result = input_string[index:]
-        print("Invalid JSON string!")
 
-    return json_obj
+            parsed_string = result.replace("\"", "\\\"")
+            pattern = r"(?<!\w)'|'(?!\w)"
+            parsed_string = re.sub(pattern, '"', parsed_string)
+            parsed_string = parsed_string.replace("\\\"", "'")
+            parsed_string = parsed_string.replace("\n\n", " ")
+
+            json_obj = json.loads(parsed_string)
+            return json_obj
+        except json.JSONDecodeError:
+            print("Invalid JSON string!")
+    else:
+        return input_string
 
 def check_fields(obj, fields):
     return all(field in obj for field in fields)
@@ -42,7 +53,7 @@ def make_openai_call(messages, token_count, fields_to_check, temperature):
     processed_response = process_response(result["choices"][0]["message"]["content"])
     if check_fields(processed_response, fields_to_check) is False and try_count < TRY_LIMIT:
         try_count = try_count + 1
-        return make_openai_call(messages, token_count, fields_to_check)
+        return make_openai_call(messages, token_count, fields_to_check, temperature)
     elif try_count >= TRY_LIMIT:
         try_count = 0
         return result["choices"][0]["message"]["content"]

helper/speech_to_text_helper.py

@@ -0,0 +1,11 @@
+import whisper
+import os
+
+def speech_to_text(file_path):
+    if os.path.exists(file_path):
+        model = whisper.load_model("base")
+        result = model.transcribe(file_path, fp16=False, language='English', verbose=False)
+        return result["text"]
+    else:
+        print("File not found:", file_path)
+        raise Exception("File " + file_path + " not found.")

sp1_playground.py

@@ -38,18 +38,14 @@ def correct_answer(
                     "content": f"The question you have to grade is of type {question_type} and is the following: {question}",
                 },
                 {
-                    "role": "system",
+                    "role": "user",
                     "content": "Please provide a JSON object response with the overall grade and breakdown grades, "
                                "formatted as follows: {'overall': 7.0, 'task_response': {'Fluency and Coherence': 8.0, "
                                "'Lexical Resource': 6.5, 'Grammatical Range and Accuracy': 7.5, 'Pronunciation': "
                                "6.0}}",
                 },
                 {
-                    "role": "system",
+                    "role": "user",
-                    "content": "Don't give explanations for the grades, just provide the json with the grades.",
-                },
-                {
-                    "role": "system",
                     "content": "If the answer is unrelated to the question give it the minimum grade.",
                 },
                 {