Initial APP

Dockerfile

@@ -0,0 +1,14 @@
+FROM python:3.9
+
+RUN useradd -m -u 1000 user
+USER user
+ENV PATH="/home/user/.local/bin:$PATH"
+
+WORKDIR /app
+
+COPY --chown=user ./requirements.txt requirements.txt
+RUN pip install --no-cache-dir --upgrade -r requirements.txt
+
+COPY --chown=user . /app
+
+CMD ["gunicorn", "main:app", "-k", "uvicorn.workers.UvicornWorker", "--bind", "0.0.0.0:7860", "--workers", "2"]

app/main.py

@@ -0,0 +1,32 @@
+from fastapi import FastAPI
+from app.routes.nlp import load_model_nlp, router as nlp_router
+from app.routes.tts import load_model_tts
+from app.routes.asr import load_model_asr, router as asr_router
+from app.routes.translation import load_model_translation, router as trans_router
+import os
+
+# Initialize application
+app = FastAPI(debug=False)
+
+# Load the pre-trained NLP 
+app.state.model_nlp, app.state.tokenizer_nlp = load_model_nlp()
+
+# Load the pre-trained Translation
+app.state.model_trans, app.state.tokenizer_trans = load_model_translation()
+
+# Load the pre-trained TTS 
+app.state.model_tts = load_model_tts()
+
+# Load the pre-trained ASR
+app.state.processor_asr, app.state.model_asr = load_model_asr()
+
+# Include the NLP router
+app.include_router(nlp_router)
+# Include the translation router
+app.include_router(trans_router)
+# Include the ASR router
+app.include_router(asr_router)
+
+@app.get("/")
+def root():
+    return {"message": "Welcome to the English Learning API"}

app/routes/asr.py

@@ -0,0 +1,36 @@
+from transformers import WhisperForConditionalGeneration, WhisperProcessor
+from fastapi import APIRouter, Request, UploadFile, File
+import librosa
+import os
+
+router = APIRouter()
+
+def load_model_asr():
+    processor = WhisperProcessor.from_pretrained("openai/whisper-small.en")
+    model = WhisperForConditionalGeneration.from_pretrained("openai/whisper-small.en")
+    return processor, model
+
+@router.post("/asr")
+async def asr(request: Request, audio: UploadFile = File(...)):
+    # Get the loaded ASR model and processor
+    processor, model = request.app.state.processor_asr, request.app.state.model_asr
+    # Audio file path
+    audio_path = os.path.join(request.app.state.AUDIO_DIR, "temp", audio.filename)
+    with open(audio_path, "wb") as f:
+        f.write(await audio.read())
+
+    # Loading audio file 
+    audio_data, sampling_rate = librosa.load(audio_path, sr=16000)
+
+    # Preparing input data
+    inputs = processor(audio_data, return_tensors="pt", sampling_rate=sampling_rate)
+    input_features = inputs["input_features"]
+
+    # Generating token IDs
+    output = model.generate(input_features)
+
+    # Decoding tokens into text
+    transcription = processor.batch_decode(output, skip_special_tokens=True)
+
+
+    return {"transcription": transcription[0]}

app/routes/nlp.py

@@ -0,0 +1,126 @@
+from transformers import AutoModelForCausalLM, AutoTokenizer
+from pydantic import BaseModel
+from fastapi import APIRouter, Request
+from .tts import save_audio
+
+
+# Model name for NLP
+model_name = "Qwen/Qwen2.5-1.5B-Instruct"
+router = APIRouter()
+
+class ChatRequest(BaseModel):
+    message: str
+    
+# Load NLP model and tokenizer
+def load_model_nlp():
+    model = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype="auto", device_map="auto")
+    tokenizer = AutoTokenizer.from_pretrained(model_name)
+    return model, tokenizer
+
+# Handle chat requests
+@router.post("/chat")
+async def chat(request: Request, message: ChatRequest):
+    message = message.message
+    # Get the loaded NLP model and tokenizer
+    model, tokenizer = request.app.state.model_nlp, request.app.state.tokenizer_nlp
+    
+    # Prepare the conversation context
+    messages = [
+        {"role": "system", "content": """
+You are Emma — a friendly, patient, encouraging native speaker of American English and an experienced English teacher. Assume every user is learning English.
+
+Top priorities (in order):
+
+First: Reply NATURALLY and CONVERSATIONALLY to the user’s most recent (last) message. The reply should sound like a warm, helpful human: concise (2–4 sentences), encouraging, and easy to understand.
+
+Second: Immediately after that natural reply, analyze only that same most recent message for language errors and apply the correction rules below. Do not analyze earlier messages.
+
+What to detect (error categories):
+
+Grammar (tenses, word order, auxiliary duplication like “what’s is”, subject-verb agreement)
+
+Vocabulary (word choice, false friends, awkward collocations)
+
+Spelling
+
+Punctuation
+
+Register (formal vs. informal mismatch)
+
+Typical learner errors (missing articles, capitalization mistakes, double auxiliaries, common typos)
+
+Correction rules:
+
+If any errors are found, append exactly one correction block at the end of your reply. If no errors are found, append nothing.
+
+Corrections must be concise, clear, encouraging, and not overwhelming.
+
+Explanations must be one sentence and simple.
+
+Provide an example only if helpful, and keep it short (one sentence).
+
+If multiple possible fixes exist, show the single most natural and simple correction for the learner (you may include a second only if it’s essential).
+
+Exact correction block format (use this format verbatim):
+
+CORRECTION:
+
+Error: [short label — e.g. “Grammar” / “Spelling” / “Vocabulary”]
+
+Original: “...original text fragment...”
+
+Correction: “...suggested correction...”
+
+Explanation: [one-sentence, simple explanation]
+(If helpful) Example: “...full correct sentence...”
+
+Behavior & style constraints:
+
+Always prioritize the conversational reply above the correction. The correction is an add-on, never the primary content.
+
+Tone: friendly, supportive, patient, non-judgmental.
+
+Keep everything short, organized, and easy to scan.
+
+Never invent facts. If you don’t know something, say “I don’t know” or ask a clarifying question.
+
+Assume the user is an English learner and tailor explanations accordingly.
+
+No long grammar essays; keep corrections short and actionable.
+
+Execution notes for the model (internal-use guidance you should follow):
+
+Analyze only the last user message text (no earlier context).
+
+If the last message contains more than one error, include up to two prioritized corrections inside the single correction block (choose the two most important).
+
+Use natural, learner-friendly wording in explanations.
+
+Keep the correction block compact and visually distinct from the conversational reply.
+
+Use your prompt-optimization and code-writing strengths to keep instructions minimal but robust — be decisive and pick the clearest fix.
+
+Final instruction: Reply to the user’s most recent message now, following these rules exactly.
+"""},
+        {"role": "user", "content": message},
+    ]
+    
+    # Tokenize input and generate a response
+    text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
+    model_inputs = tokenizer([text], return_tensors="pt").to(model.device)
+    generated_ids = model.generate(
+        **model_inputs, 
+        max_new_tokens=512,
+        top_p=0.9,
+        temperature=0.7,
+        do_sample=True,
+        pad_token_id=tokenizer.eos_token_id)
+    
+    # Decode the response
+    generated_ids = [output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)]
+    response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
+    
+    # Save response as audio
+    url_path = save_audio(request, response)
+
+    return {"response": response, "audio": url_path}

app/routes/translation.py

@@ -0,0 +1,29 @@
+from transformers import AutoModelForSeq2SeqLM, AutoTokenizer
+from fastapi import APIRouter, Request
+from pydantic import BaseModel
+
+model_name = "allegro/BiDi-eng-pol"
+router = APIRouter()
+
+class TextInput(BaseModel):
+    text: str 
+
+# Ładowanie modelu tłumaczenia
+def load_model_translation():
+    tokenizer = AutoTokenizer.from_pretrained(model_name)
+    model = AutoModelForSeq2SeqLM.from_pretrained(model_name)
+    return model, tokenizer
+
+@router.post("/translate")
+async def translate_text(request: Request, text: TextInput):
+    model, tokenizer = request.app.state.model_trans, request.app.state.tokenizer_trans
+
+    # Prefiks >>pol<< informuje model, że ma tłumaczyć na polski
+    text = ">>pol<< " + text.text
+
+    # Tokenizacja i generowanie tłumaczenia
+    inputs = tokenizer([text], return_tensors="pt", padding=True)
+    translated = model.generate(**inputs)
+    decoded_translation = tokenizer.decode(translated[0], skip_special_tokens=True)
+
+    return {"translation": decoded_translation}

app/routes/tts.py

@@ -0,0 +1,28 @@
+from fastapi import Request
+from kokoro import KPipeline
+import numpy as np
+import soundfile as sf
+import os
+import uuid
+
+# Ładowanie modelu Kokoro tylko raz przy starcie aplikacji
+def load_model_tts():
+    pipeline = KPipeline(lang_code='a')  # 'a' = automatyczne wykrycie języka
+    return pipeline
+
+# Funkcja generująca i zapisująca audio
+def save_audio(request: Request, text: str, voice: str = 'af_heart'):
+    pipeline = request.app.state.model_tts
+
+    file_name = f"{uuid.uuid4()}.wav"
+    file_path = os.path.join(request.app.state.AUDIO_DIR, file_name)
+    # Initialize an empty array to merge all audio fragments
+    audio_total = np.array([], dtype=np.float32)
+    
+    # We generate audio in streaming mode (the generator returns fragments) 
+    generator = pipeline(text, voice=voice)
+    for _, _, audio in generator:
+        audio_total = np.concatenate([audio_total, audio])
+
+    sf.write(file_path, audio_total, 24000)
+    return f"http://127.0.0.1:8000/audio/{file_name}"

requirements.txt

@@ -0,0 +1,197 @@
+absl-py==2.1.0
+accelerate==1.5.2
+addict==2.4.0
+aiohappyeyeballs==2.6.1
+aiohttp==3.11.13
+aiosignal==1.3.2
+annotated-types==0.7.0
+anyio==4.9.0
+asttokens==3.0.0
+astunparse==1.6.3
+attrs==25.3.0
+audioread==3.0.1
+babel==2.17.0
+blis==1.2.0
+catalogue==2.0.10
+certifi==2025.1.31
+cffi==1.17.1
+charset-normalizer==3.4.1
+click==8.1.8
+cloudpathlib==0.21.0
+colorama==0.4.6
+confection==0.1.5
+csvw==3.5.1
+curated-tokenizers==0.0.9
+curated-transformers==0.1.1
+cymem==2.0.11
+datasets==3.4.0
+decorator==5.2.1
+dill==0.3.8
+Distance==0.1.3
+dlinfo==2.0.0
+dnspython==2.7.0
+docopt==0.6.2
+email_validator==2.2.0
+en_core_web_sm @ https://github.com/explosion/spacy-models/releases/download/en_core_web_sm-3.8.0/en_core_web_sm-3.8.0-py3-none-any.whl#sha256=1932429db727d4bff3deed6b34cfc05df17794f4a52eeb26cf8928f7c1a0fb85
+espeakng-loader==0.2.4
+executing==2.2.0
+fastapi==0.115.11
+fastapi-cli==0.0.7
+filelock==3.18.0
+flatbuffers==25.2.10
+frozenlist==1.5.0
+fsspec==2024.12.0
+g2p-en==2.1.0
+gast==0.6.0
+google-pasta==0.2.0
+grpcio==1.71.0
+h11==0.14.0
+h5py==3.13.0
+httpcore==1.0.7
+httptools==0.6.4
+httpx==0.28.1
+huggingface-hub==0.29.3
+idna==3.10
+inflect==7.5.0
+ipython==9.0.2
+ipython_pygments_lexers==1.1.1
+isodate==0.7.2
+itsdangerous==2.2.0
+jedi==0.19.2
+Jinja2==3.1.6
+joblib==1.4.2
+jsonschema==4.23.0
+jsonschema-specifications==2024.10.1
+keras==3.9.0
+kokoro==0.9.4
+langcodes==3.5.0
+language-tags==1.2.0
+language_data==1.3.0
+lazy_loader==0.4
+libclang==18.1.1
+librosa==0.11.0
+llvmlite==0.44.0
+loguru==0.7.3
+marisa-trie==1.2.1
+Markdown==3.7
+markdown-it-py==3.0.0
+MarkupSafe==3.0.2
+matplotlib-inline==0.1.7
+mdurl==0.1.2
+misaki==0.9.4
+ml_dtypes==0.5.1
+more-itertools==10.6.0
+mpmath==1.3.0
+msgpack==1.1.0
+multidict==6.1.0
+multiprocess==0.70.16
+murmurhash==1.0.12
+namex==0.0.8
+networkx==3.4.2
+nltk==3.9.1
+num2words==0.5.14
+numba==0.61.0
+numpy==1.26.4
+nvidia-cublas-cu12==12.4.5.8
+nvidia-cuda-cupti-cu12==12.4.127
+nvidia-cuda-nvrtc-cu12==12.4.127
+nvidia-cuda-runtime-cu12==12.4.127
+nvidia-cudnn-cu12==9.1.0.70
+nvidia-cufft-cu12==11.2.1.3
+nvidia-curand-cu12==10.3.5.147
+nvidia-cusolver-cu12==11.6.1.9
+nvidia-cusparse-cu12==12.3.1.170
+nvidia-cusparselt-cu12==0.6.2
+nvidia-nccl-cu12==2.21.5
+nvidia-nvjitlink-cu12==12.4.127
+nvidia-nvtx-cu12==12.4.127
+opt_einsum==3.4.0
+optree==0.14.1
+orjson==3.10.15
+packaging==24.2
+pandas==2.2.3
+parso==0.8.4
+pexpect==4.9.0
+phonemizer-fork==3.3.2
+platformdirs==4.3.6
+pooch==1.8.2
+preshed==3.0.9
+prompt_toolkit==3.0.50
+propcache==0.3.0
+protobuf==5.29.3
+psutil==7.0.0
+ptyprocess==0.7.0
+pure_eval==0.2.3
+pyarrow==19.0.1
+pycparser==2.22
+pydantic==2.10.6
+pydantic-extra-types==2.10.3
+pydantic-settings==2.8.1
+pydantic_core==2.27.2
+Pygments==2.19.1
+pyparsing==3.2.1
+python-dateutil==2.9.0.post0
+python-dotenv==1.0.1
+python-multipart==0.0.20
+pytz==2025.1
+PyYAML==6.0.2
+rdflib==7.1.3
+referencing==0.36.2
+regex==2024.11.6
+requests==2.32.3
+rfc3986==1.5.0
+rich==13.9.4
+rich-toolkit==0.13.2
+rpds-py==0.23.1
+safetensors==0.5.3
+scikit-learn==1.6.1
+scipy==1.15.2
+segments==2.3.0
+sentencepiece==0.2.0
+setuptools==76.0.0
+shellingham==1.5.4
+six==1.17.0
+smart-open==7.1.0
+sniffio==1.3.1
+soundfile==0.13.1
+soxr==0.5.0.post1
+spacy==3.8.4
+spacy-curated-transformers==0.3.0
+spacy-legacy==3.0.12
+spacy-loggers==1.0.5
+srsly==2.5.1
+stack-data==0.6.3
+starlette==0.46.1
+sympy==1.13.1
+tensorboard==2.19.0
+tensorboard-data-server==0.7.2
+tensorflow==2.19.0
+termcolor==2.5.0
+thinc==8.3.4
+threadpoolctl==3.6.0
+tokenizers==0.21.1
+torch==2.6.0
+torchaudio==2.6.0
+tqdm==4.67.1
+traitlets==5.14.3
+transformers==4.49.0
+triton==3.2.0
+typeguard==4.4.2
+typer==0.15.2
+typing_extensions==4.12.2
+tzdata==2025.1
+ujson==5.10.0
+uritemplate==4.1.1
+urllib3==2.3.0
+uvicorn==0.34.0
+uvloop==0.21.0
+wasabi==1.1.3
+watchfiles==1.0.4
+wcwidth==0.2.13
+weasel==0.4.1
+websockets==15.0.1
+Werkzeug==3.1.3
+wheel==0.45.1
+wrapt==1.17.2
+xxhash==3.5.0
+yarl==1.18.3