new structure for transcription engine

- found problems with mps
- integrated jinja2 templates for rendering (.tex?)
- raw ui & bad structure

.gitignore

@@ -10,4 +10,7 @@ wheels/
 .venv
 
 main.todo
-sample.mp3
+sample.mp3
+*.spec
+
+ui/assets/question_mark.png

README.md

@@ -2,9 +2,14 @@
 
 **Notecast** is a simple and minimalistic utilite to transcribe audiofiles to text and create a conspect.
 
-Requires nvcc v12.8
+Requires nvidia cuda toolkit v12.4
 
 You can create environment running
 ```
 conda env create -f environment.yml
+```
+or for CUDA/Nvidia:
+```
+conda create -n notecast -c pytorch -c nvidia pytorch torchvision torchaudio transformers python=3.12
+conda install ffmpeg customtkinter openai -c conda-forge -c bioconda
 ```

env.yml

@@ -0,0 +1,15 @@
+name: notecast-cpu-mps
+channels:
+  - pytorch
+  - conda-forge
+  - defaults
+  - bioconda
+dependencies:
+  - pytorch
+  - ffmpeg
+  - torchvision
+  - torchaudio
+  - transformers
+  - python=3.12
+  - customtkinter
+  - openai

env_cuda.yml

@@ -0,0 +1,15 @@
+name: notecast-cuda
+channels:
+  - pytorch
+  - nvidia
+  - conda-forge
+  - bioconda
+dependencies:
+  - pytorch
+  - ffmpeg
+  - torchvision
+  - torchaudio
+  - transformers
+  - python=3.12
+  - customtkinter
+  - openai

environment.yml

@@ -1,195 +0,0 @@
-name: notecast
-channels:
-  - conda-forge
-  - defaults
-  - nvidia
-  - pytorch
-dependencies:
-  - accelerate=1.10.1=pyhcf101f3_0
-  - aiohappyeyeballs=2.6.1=pyhd8ed1ab_0
-  - aiohttp=3.12.15=pyh7db6752_0
-  - aiosignal=1.4.0=pyhd8ed1ab_0
-  - aom=3.6.0=hd77b12b_0
-  - arrow-cpp=19.0.0=h82e8c66_4
-  - async-timeout=5.0.1=pyhd8ed1ab_1
-  - attrs=25.3.0=pyh71513ae_0
-  - aws-c-auth=0.9.0=hd490b63_15
-  - aws-c-cal=0.9.2=hd8a8e38_0
-  - aws-c-common=0.12.3=h2466b09_0
-  - aws-c-compression=0.3.1=h5d0e663_5
-  - aws-c-event-stream=0.5.5=ha416645_0
-  - aws-c-http=0.10.2=h81282ae_2
-  - aws-c-io=0.20.1=hf7624bd_1
-  - aws-c-mqtt=0.13.1=h5c1ae27_3
-  - aws-c-s3=0.8.3=h1e843c7_0
-  - aws-c-sdkutils=0.2.4=h5d0e663_0
-  - aws-checksums=0.2.7=h5d0e663_1
-  - aws-crt-cpp=0.32.10=h8abd1a4_2
-  - aws-sdk-cpp=1.11.528=hd293286_1
-  - blas=1.0=mkl
-  - bottleneck=1.4.2=py312h4b0e54e_0
-  - brotlicffi=1.0.9.2=py312h5da7b33_1
-  - bzip2=1.0.8=h2bbff1b_6
-  - c-ares=1.34.5=h2466b09_0
-  - ca-certificates=2025.8.3=h4c7d964_0
-  - cairo=1.18.4=he9e932c_0
-  - certifi=2025.8.3=py312haa95532_0
-  - cffi=1.17.1=py312h827c3e9_1
-  - charset-normalizer=3.3.2=pyhd3eb1b0_0
-  - colorama=0.4.6=pyhd8ed1ab_1
-  - cuda-cccl=13.0.50=h23517cc_1
-  - cuda-cccl_win-64=13.0.50=hc667259_1
-  - cuda-cudart=12.4.127=0
-  - cuda-cudart-dev=12.4.127=0
-  - cuda-cupti=12.4.127=0
-  - cuda-libraries=12.4.1=0
-  - cuda-libraries-dev=12.4.1=0
-  - cuda-nvrtc=12.4.127=0
-  - cuda-nvrtc-dev=12.4.127=0
-  - cuda-nvtx=12.4.127=0
-  - cuda-opencl=13.0.39=0
-  - cuda-opencl-dev=13.0.39=0
-  - cuda-profiler-api=13.0.39=0
-  - cuda-runtime=12.4.1=0
-  - cuda-version=13.0=3
-  - datasets=4.0.0=pyhcf101f3_0
-  - dav1d=1.2.1=h2bbff1b_0
-  - dill=0.3.8=pyhd8ed1ab_0
-  - expat=2.7.1=h8ddb27b_0
-  - ffmpeg=4.3.1=ha925a31_0
-  - filelock=3.17.0=py312haa95532_0
-  - fontconfig=2.14.1=hb33846d_3
-  - freeglut=3.4.0=h8a1e904_1
-  - freetype=2.13.3=h0620614_0
-  - fribidi=1.0.10=h62dcd97_0
-  - frozenlist=1.7.0=pyhf298e5d_0
-  - fsspec=2025.3.0=pyhd8ed1ab_0
-  - gflags=2.2.2=he0c23c2_1005
-  - giflib=5.2.2=h7edc060_0
-  - glog=0.5.0=h4797de2_0
-  - gmp=6.3.0=h537511b_0
-  - gmpy2=2.2.1=py312h827c3e9_0
-  - graphite2=1.3.14=hd77b12b_1
-  - harfbuzz=10.2.0=he2f9f60_1
-  - hf-xet=1.1.8=py312h79d111c_0
-  - huggingface_hub=0.34.4=pyhd8ed1ab_1
-  - icu=73.1=h6c2663c_0
-  - idna=3.7=py312haa95532_0
-  - intel-openmp=2023.1.0=h59b6b97_46320
-  - jinja2=3.1.6=py312haa95532_0
-  - jpeg=9e=h827c3e9_3
-  - khronos-opencl-icd-loader=2024.05.08=h8cc25b3_0
-  - lcms2=2.16=h62be587_1
-  - lerc=4.0.0=h5da7b33_0
-  - libabseil=20250127.0=cxx17_h4eb7d71_0
-  - libavif=1.1.1=h827c3e9_0
-  - libbrotlicommon=1.0.9=hcfcfb64_9
-  - libbrotlidec=1.0.9=hcfcfb64_9
-  - libbrotlienc=1.0.9=hcfcfb64_9
-  - libcublas=12.4.5.8=0
-  - libcublas-dev=12.4.5.8=0
-  - libcufft=11.2.1.3=0
-  - libcufft-dev=11.2.1.3=0
-  - libcurand=10.4.0.35=0
-  - libcurand-dev=10.4.0.35=0
-  - libcurl=8.15.0=h2300eb9_0
-  - libcusolver=11.6.1.9=0
-  - libcusolver-dev=11.6.1.9=0
-  - libcusparse=12.3.1.170=0
-  - libcusparse-dev=12.3.1.170=0
-  - libdeflate=1.22=h5bf469e_0
-  - libffi=3.4.4=hd77b12b_1
-  - libglib=2.84.2=h405b238_0
-  - libgrpc=1.71.0=hf4237ab_0
-  - libiconv=1.16=h2bbff1b_3
-  - libjpeg-turbo=2.0.0=h196d8e1_0
-  - libnpp=12.2.5.30=0
-  - libnpp-dev=12.2.5.30=0
-  - libnvfatbin=13.0.39=0
-  - libnvfatbin-dev=13.0.39=0
-  - libnvjitlink=12.4.127=0
-  - libnvjitlink-dev=12.4.127=0
-  - libnvjpeg=12.3.1.117=0
-  - libnvjpeg-dev=12.3.1.117=0
-  - libpng=1.6.39=h8cc25b3_0
-  - libprotobuf=5.29.3=h65a231f_1
-  - libre2-11=2024.07.02=hd248061_3
-  - libssh2=1.11.1=h2addb87_0
-  - libthrift=0.15.0=ha2884a9_3
-  - libtiff=4.7.0=h404307b_0
-  - libuv=1.48.0=h827c3e9_0
-  - libwebp=1.3.2=h18467be_1
-  - libwebp-base=1.3.2=h3d04722_1
-  - libxml2=2.13.8=h866ff63_0
-  - lz4-c=1.9.4=h2bbff1b_1
-  - markupsafe=3.0.2=py312h827c3e9_0
-  - mkl=2023.1.0=h6b88ed4_46358
-  - mkl-service=2.4.0=py312h827c3e9_2
-  - mkl_fft=1.3.11=py312h827c3e9_0
-  - mkl_random=1.2.8=py312h0158946_0
-  - mpc=1.3.1=h827c3e9_0
-  - mpfr=4.2.1=h56c3642_0
-  - mpmath=1.3.0=py312haa95532_0
-  - multidict=6.6.3=pyh62beb40_0
-  - multiprocess=0.70.15=py312haa95532_0
-  - networkx=3.5=py312haa95532_0
-  - numexpr=2.11.0=py312hdb065b2_0
-  - numpy=2.3.1=py312h5f75535_0
-  - numpy-base=2.3.1=py312h23d94f8_0
-  - openjpeg=2.5.2=h9b5d1b5_1
-  - openssl=3.5.2=h725018a_0
-  - orc=2.1.1=hd1c1d5c_0
-  - packaging=25.0=pyh29332c3_1
-  - pandas=2.3.2=py312ha5e6156_0
-  - pcre2=10.42=h0ff8eda_1
-  - pillow=11.3.0=py312hb328d1f_0
-  - pip=25.2=pyhc872135_0
-  - pixman=0.46.4=h4043f72_0
-  - propcache=0.3.1=pyhe1237c8_0
-  - psutil=5.9.0=py312h827c3e9_1
-  - pyarrow=19.0.0=py312h5da7b33_1
-  - pycparser=2.21=pyhd3eb1b0_0
-  - pysocks=1.7.1=py312haa95532_0
-  - python=3.12.11=h716150d_0
-  - python-dateutil=2.9.0.post0=pyhe01879c_2
-  - python-tzdata=2025.2=pyhd8ed1ab_0
-  - python-xxhash=3.5.0=py312h827c3e9_0
-  - pytorch=2.5.1=py3.12_cuda12.4_cudnn9_0
-  - pytorch-cuda=12.4=h3fd98bf_7
-  - pytorch-mutex=1.0=cuda
-  - pytz=2025.2=pyhd8ed1ab_0
-  - pyyaml=6.0.2=py312h827c3e9_0
-  - re2=2024.07.02=haf4117d_3
-  - regex=2024.11.6=py312h827c3e9_0
-  - requests=2.32.5=py312haa95532_0
-  - safetensors=0.5.3=py312h44068b5_0
-  - setuptools=72.1.0=py312haa95532_0
-  - six=1.17.0=pyhe01879c_1
-  - snappy=1.2.2=h7fa0ca8_0
-  - sqlite=3.50.2=hda9a48d_1
-  - sympy=1.14.0=py312haa95532_0
-  - tbb=2021.8.0=h59b6b97_0
-  - tk=8.6.15=hf199647_0
-  - tokenizers=0.21.0=py312h482ea96_0
-  - tqdm=4.67.1=pyhd8ed1ab_1
-  - transformers=4.55.4=pyhd8ed1ab_0
-  - typing-extensions=4.15.0=py312haa95532_0
-  - typing_extensions=4.15.0=py312haa95532_0
-  - tzdata=2025b=h04d1e81_0
-  - ucrt=10.0.22621.0=haa95532_0
-  - urllib3=2.5.0=py312haa95532_0
-  - utf8proc=2.6.1=h2bbff1b_1
-  - vc=14.42=haa95532_5
-  - vc14_runtime=14.44.35208=h4927774_10
-  - vs2015_runtime=14.44.35208=ha6b5a95_10
-  - wheel=0.45.1=py312haa95532_0
-  - win_inet_pton=1.1.0=py312haa95532_0
-  - xxhash=0.8.0=h8ffe710_3
-  - xz=5.6.4=h4754444_1
-  - yaml=0.2.5=he774522_0
-  - yarl=1.20.1=pyhe1237c8_0
-  - zlib=1.2.13=h8cc25b3_1
-  - zstd=1.5.6=h8880b57_0
-  - pip:
-      - torchaudio==2.5.1
-      - torchvision==0.20.1

main.py

@@ -1,13 +1,5 @@
-from transcription.audio_transcription import AudioTranscription
+from ui.ui import TranscriberApp
-from transcription.torch_checker import check_torch
-from utils.logger import setup_logger
 
-logger = setup_logger("main")
+if __name__ == "__main__":
-
+    app = TranscriberApp()
-check_torch()
+    app.mainloop()
-
-try:
-    track = AudioTranscription("sample.mp3")
-    print(track.transcribe_audio())
-except Exception as e:
-    logger.error(f"Execution error: {e}") 

make_venv.ps1

@@ -1,13 +0,0 @@
-$envName = "notecast"
-
-Write-Output >>> Creating environment $envName from environment.yml"
-conda env create -f environment.yml -n $envName
-if ($LASTEXITCODE -ne 0) {
-    Write-Output ">>> Environment already exists, updating..."
-    conda env update -f environment.yml --prune
-}
-
-Write-Output ">>> Activating environment"
-conda activate $envName
-
-Write-Output ">>> Download completed!"

make_venv.sh

@@ -1,13 +0,0 @@
-#!/bin/bash
-set -e
-
-ENV_NAME="notecast"
-
-echo ">>> Creating environment $ENV_NAME from environment.yml"
-conda env create -f environment.yml || conda env update -f environment.yml --prune
-
-echo ">>> Activating environment"
-eval "$(conda shell.bash hook)"
-conda activate "$ENV_NAME"
-
-echo ">>> Download completed!"

transcription/audio.py

@@ -0,0 +1,17 @@
+from dataclasses import dataclass
+import torchaudio
+import torch
+
+class Audio:
+    waveform: torch.Tensor
+    sr: int
+
+    def load(self, filepath):
+        """
+        Loads audio from file's path
+        """
+        self.waveform, self.sr = torchaudio.load(
+            filepath,
+            backend="ffmpeg",
+        )
+        return self

transcription/audio_transcription.py

@@ -1,143 +1,50 @@
-from transformers import WhisperProcessor, WhisperForConditionalGeneration
+from transcription.audio import Audio
-import torch
+from transcription.preprocessing.audio_preprocessor import AudioPreprocessor
-import torchaudio
+from transcription.preprocessing.splitter import Splitter
-from utils.logger import setup_logger
+from transcription.engines.whisper import WhisperEngine
-import time
+from transcription.configuration import Configuration
-import math
-from tqdm import tqdm
-
-logger = setup_logger("AudioTranscribe module")
 
+# maybe inherit from AudioTranscription and rename to something like WhisperTranscription?
 class AudioTranscription:
-    model_name = "openai/whisper-large-v2" 
+    # add multimodel ability
-    
-    filepath: str
-    waveform: torch.Tensor
-    sampling_rate: int
-    
-    chunks: list = []
-    chunk_size: int 
-    device = "cuda"
-    processor: WhisperProcessor
-    model: WhisperForConditionalGeneration
-    
-    language = "ru"
-    
-    all_transcription: list = []
-    
     def __init__(
-        self, 
+        self,
         filepath: str,
-        language = "ru",
+        config: Configuration,
-        device = "cuda",
+        language,
-        model_name = "openai/whisper-large-v2"
+        # logger
     ) -> None:
         self.filepath = filepath
         self.language = language
-        self.device = device
+        # self.logger = logger
-        self.model_name = model_name
+
-        self.chunks: list = []
+        self.audio = Audio()
-        try:
+        self.preprocessor = AudioPreprocessor()
-            logger.info("Loading model WhisperProcessor...")
+        self.splitter = Splitter(
-            self.processor = WhisperProcessor.from_pretrained(self.model_name)
+            chunkSize=config.chunkSize,
-            
+            batchSize=config.batchSize,
-            self.model = WhisperForConditionalGeneration.from_pretrained(
+        )
-                self.model_name,
+        self.engine = WhisperEngine(
-                torch_dtype=torch.float16,
+            modelName=config.modelName,
-                device_map="auto"
+            language=self.language,
-            ).to(self.device)
+            dType=config.dType,
-            
+            device=config.device,
-            logger.info("Model loaded.")
+        )
-            
+    
-            self.waveform, self.sampling_rate = torchaudio.load(filepath, format="mp3", backend="ffmpeg")
+    # maybe add something like temperature here?
-            logger.info(f"Successfully loaded file {filepath}.")
+    def transcribeAudio(self) -> str:
+        transcription: list = []
         
-        except Exception as e:
+        self.engine.loadModel()
-            logger.error(f"Unable to load file {self.filepath}: {e}")
-            raise
-    
-    def resample(self) -> None:
-        self.waveform = torchaudio.functional.resample(self.waveform, self.sampling_rate, 16000)
-    
-    def to_mono(self):
-        if self.waveform.shape[0] > 1:
-            self.waveform = self.waveform.mean(dim=0, keepdim=True)
-        self.waveform = self.waveform.squeeze(0)
-    
-    def split_to_chunks(self, chunk_length_s: int = 30) -> None:
-        logger.info(f"Splitting audio on chunks...")
         
-        self.chunk_size = chunk_length_s * 16000  # 16kHz after resampling
+        self.preprocessor.prepare(self.audio.load(self.filepath))
-        total_samples = self.waveform.shape[0]
+
-        chunks_count = (total_samples + self.chunk_size - 1) // self.chunk_size
+        batches = self.splitter.split(self.audio.waveform)
+
+        for batch in batches:
+            batchText: str = self.engine.transcribeBatch(batch)
+            transcription.append(batchText)
+
+        self.engine.unloadModel()
         
-        logger.info(f"File length - {total_samples / 16000:.1f} seconds, splitting on {chunks_count} chunks by {chunk_length_s} seconds per chunk.")
+        return str(" ".join(transcription))
-
-        self.chunks = []
-        for idx in range(chunks_count):
-            start = idx * self.chunk_size
-            end = min((idx + 1) * self.chunk_size, total_samples)
-            chunk = self.waveform[start:end].cpu().numpy().astype("float32")
-            self.chunks.append(chunk)
-    
-    def process_chunk(
-        self, 
-        chunk
-    ) -> str:
-        inputs = self.processor(chunk, sampling_rate=16000, return_tensors="pt")
-        input_features = inputs.input_features.to(self.device).to(torch.float16)
-
-        with torch.no_grad():
-            predicted_ids = self.model.generate(
-                input_features,
-                language=self.language,
-                task="transcribe",
-                temperature=0.0
-            )
-        
-        text = self.processor.batch_decode(predicted_ids, skip_special_tokens=True)[0]
-        
-        return text
-    
-    def process_all_chunks(self, batch_size: int = 16) -> None:
-        start_time = time.time()
-        try:
-            self.all_transcription = []
-
-            for i in tqdm(range(math.ceil(len(self.chunks) / batch_size))):
-                # TODO: rewrite batching as a separate function
-                batch = self.chunks[i*batch_size:(i+1)*batch_size]
-
-                inputs = self.processor(
-                    batch, 
-                    sampling_rate=16000, 
-                    return_tensors="pt", 
-                    padding=True
-                )
-
-                input_features = inputs.input_features.to(self.device).to(torch.float16)
-
-                with torch.no_grad():
-                    predicted_ids = self.model.generate(
-                        input_features,
-                        language=self.language,
-                        task="transcribe",
-                        temperature=0.0
-                    )
-
-                texts = self.processor.batch_decode(predicted_ids, skip_special_tokens=True)
-                self.all_transcription.extend(texts)
-
-            end_time = time.time()
-            logger.info(f"Transcription completed in {end_time - start_time:.2f} seconds")
-
-        except Exception as e:
-            logger.error(f"Errors occured while processing chunks: {e}")
-
-    
-    def transcribe_audio(self) -> str:
-        self.resample()
-        self.to_mono()
-        self.split_to_chunks()
-        self.process_all_chunks()
-        return " ".join(self.all_transcription)

transcription/configuration.py

@@ -0,0 +1,23 @@
+from dataclasses import dataclass
+
+import torch
+
+@dataclass
+class Configuration:
+    # add new models
+    device: str = "cuda"
+    modelName: str = "openai/whisper-large-v2"
+    chunkSize: int = 30
+    batchSize: int = 16
+    dataType: str = "torch.float16"
+
+    _dtype_map = {
+        "torch.float16": torch.float16,
+        "torch.float32": torch.float32,
+        "torch.bfloat16": torch.bfloat16,
+    }
+
+    dType: torch.dtype = None
+
+    def __post_init__(self):
+        self.dType = self._dtype_map[self.dataType]

transcription/engines/base_engine.py

@@ -0,0 +1,30 @@
+import torch
+from abc import ABC, abstractmethod
+
+class BaseEngine(ABC):
+    def __init__(
+        self,
+        modelName: str,
+        language: str,
+        dType: torch.dtype,
+        device: str
+    ):
+        self.modelName = modelName
+        self.device = device
+        self.language = language
+        self.dType = dType
+    
+    @abstractmethod
+    def loadModel(self) -> None:
+        pass
+
+    @abstractmethod
+    def unloadModel(self) -> None:
+        pass
+
+    @abstractmethod
+    def transcribeBatch(
+        self, 
+        batch
+    ) -> str:
+        pass

transcription/engines/whisper.py

@@ -0,0 +1,65 @@
+# from logging import Logger
+import time
+import torch
+import gc
+from transformers import WhisperForConditionalGeneration, WhisperProcessor
+
+from transcription.engines.base_engine import BaseEngine
+
+class WhisperEngine(BaseEngine):
+    def loadModel(self) -> None:
+        self.processor = WhisperProcessor.from_pretrained(self.modelName)
+        self.model = WhisperForConditionalGeneration.from_pretrained(
+            self.modelName,
+            torch_dtype = self.dType # check twice
+        ).to(self.device) # ??? recheck
+    
+    def unloadModel(self) -> None:
+        self.model = None
+        self.processor = None
+
+        # TODO: MPS?
+        if self.device == "cuda": 
+            torch.cuda.empty_cache()
+    
+    def transcribeBatch(
+        self,
+        batch,
+    ) -> str:
+        assert self.processor is not None
+        assert self.model is not None
+
+        inputs = self.processor(
+            batch, 
+            sampling_rate=16000,
+            return_tensors="pt",
+            padding=True,
+        )
+        
+        input_features = inputs.input_features.to(self.device).to(self.dType)
+
+        with torch.no_grad():
+            predicted_ids = self.model.generate(
+                input_features,
+                language=self.language,
+                task="transcribe",
+                temperature=0.0,
+            )
+        
+        batchText = self.processor.batch_decode(
+            predicted_ids, 
+            skip_special_tokens=True,
+        )
+
+        inputs = None
+        input_features = None
+        predicted_ids = None
+        gc.collect()
+
+        # maybe do here something with MPS?
+        if self.device.startswith("cuda"):
+            torch.cuda.empty_cache()
+            torch.cuda.ipc_collect()
+
+        return " ".join(batchText)
+        

transcription/preprocessing/audio_preprocessor.py

@@ -0,0 +1,35 @@
+from transcription.audio import Audio
+import torchaudio
+
+class AudioPreprocessor:
+    TARGET_SAMPLING_RATE: int = 16000
+
+    # for different models in future
+    # def __init__(self, model):
+    #     pass
+
+    def _resample(
+        self,
+        audio: Audio
+    ) -> None:
+        if audio.sr != self.TARGET_SAMPLING_RATE:
+            audio.waveform = torchaudio.functional.resample(
+                audio.waveform, 
+                audio.sr, 
+                self.TARGET_SAMPLING_RATE
+            )
+    
+    def _to_mono(
+        self,
+        audio: Audio
+    ) -> None:
+        if audio.waveform.shape[0] > 1:
+            audio.waveform = audio.waveform.mean(dim=0, keepdim=True)
+        audio.waveform = audio.waveform.squeeze(0)
+    
+    def prepare(
+        self, 
+        audio: Audio
+    ):
+        self._resample(audio)
+        self._to_mono(audio)

transcription/preprocessing/splitter.py

@@ -0,0 +1,50 @@
+import torch
+from typing import List
+
+class Splitter:
+    def __init__(
+        self,
+        chunkSize: int,
+        batchSize: int,
+    ) -> None:
+        self.chunkSize = chunkSize * 16000 # 16 kHz after resampling
+        self.batchSize = batchSize
+
+    # maybe raise some exceptions here?
+    def _split_to_chunks(
+        self, 
+        waveform: torch.Tensor,
+    ) -> List:
+        totalSamples = waveform.shape[0]
+        chunksCount = (totalSamples + self.chunkSize - 1) // self.chunkSize
+
+        chunks: List = []
+        # tqdm or something here? 
+        for chunkNum in range(chunksCount):
+            start = chunkNum * self.chunkSize
+            end = min((chunkNum + 1) * self.chunkSize, totalSamples)
+
+            chunk = waveform[start : end].cpu().numpy().astype("float32")
+            chunks.append(chunk)
+        
+        return chunks
+
+    def _split_to_batches(
+        self,
+        chunks: List,
+    ) -> List:
+        batches: List = []
+
+        for i in range(0, len(chunks), self.batchSize):
+            batch = chunks[i : i + self.batchSize]
+            batches.append(batch)
+        
+        return batches
+
+    def split(
+        self,
+        waveform: torch.Tensor 
+    ) -> List:
+        chunks = self._split_to_chunks(waveform)
+        batches = self._split_to_batches(chunks)
+        return batches

transcription/torch_checker.py

@@ -1,11 +1,36 @@
+import logging
 import torch
 
-def check_torch() -> None:
+
-    print("=== Checking PyTorch ===")
+def checkTorch(logger: logging.Logger) -> None:
-    print(f"Torch version: {torch.version}")
+    logger.info("=== Checking PyTorch ===")
-    print(f"CUDA is available: {torch.cuda.is_available()}")
+    logger.info(f"Torch version: {torch.__version__}")
+
+    # NVIDIA / AMD (CUDA API)
     if torch.cuda.is_available():
-        print(f"CUDA version: {torch.version.cuda}")
+        backend = "CUDA"
-        print(f"Number of GPU: {torch.cuda.device_count()}")
+        if torch.version.hip is not None:
-        print(f"Name of GPU: {torch.cuda.get_device_name(0)}")
+            backend = "ROCm (AMD HIP)"
-    print("=== Check completed ===")
+
+        logger.info(f"{backend} backend is available")
+        logger.info(
+            f"Compiled with: CUDA {torch.version.cuda}, ROCm {torch.version.hip}"
+        )
+        logger.info(f"Number of devices: {torch.cuda.device_count()}")
+
+        for i in range(torch.cuda.device_count()):
+            logger.info(f"GPU {i}: {torch.cuda.get_device_name(i)}")
+
+    # Apple Silicon (MPS)
+    elif hasattr(torch.backends, "mps") and torch.backends.mps.is_available():
+        logger.info("MPS backend is available (Apple Silicon)")
+        logger.info(
+            f"MPS version: {getattr(torch.backends.mps, '__version__', 'unknown')}"
+        )
+        logger.info("GPU: Apple Silicon (Metal)")
+
+    # CPU only mode
+    else:
+        logger.info("Only CPU is available")
+
+    logger.info("=== Check completed ===")

ui.py

@@ -1,13 +0,0 @@
-import tkinter as tk
-
-root = tk.Tk()
-root.title("Audio Transcriptor")
-root.geometry("800x600")
-
-path_entry = tk.Entry(root, width=40)
-path_entry.pack(padx=10, pady=(5, 10))
-
-transcribe_button = tk.Button(text="Transcribe")
-transcribe_button.pack(anchor="e", rely=25, relx=15)
-
-root.mainloop()

ui/configuration_actions.py

@@ -0,0 +1,18 @@
+from transcription.configuration import Configuration
+
+
+# TODO: implement saving & removing configuration
+def save_configuration(cfg: DeviceConfiguration):
+    config = {
+        "Model": cfg.model_name,
+        "Batch Size": cfg.batch_size,
+        "Data Type": cfg.data_type,
+    }
+
+
+def load_config():
+    pass
+
+
+def delete_config():
+    pass

ui/state.py

@@ -0,0 +1,36 @@
+import tkinter as tk
+import torch
+
+class State:
+    def __init__(self, root):
+        # transcription
+        self.model = tk.StringVar(root, "openai/whisper-large-v3-turbo")
+        self.batch = tk.StringVar(root, "32")
+        self.chunk = tk.StringVar(root, "30")
+        self.dtype = tk.StringVar(root, "torch.float16")
+        self.language = tk.StringVar(root, "ru")
+
+        # llm
+        self.api_key = tk.StringVar(root, "")
+        self.api_model = tk.StringVar(root, "")
+        self.base_url = tk.StringVar(root, "")
+        self.conspect_lang = tk.StringVar(root, "Russian")
+
+        # flags
+        self.create_conspect = tk.BooleanVar(root, False)
+        self.remove_transcription = tk.BooleanVar(root, False)
+
+        # files
+        self.input_file = tk.StringVar(root)
+        self.output_file = tk.StringVar(root)
+
+        # device
+        devices = []
+        if torch.cuda.is_available():
+            devices.append("cuda")
+        if torch.backends.mps.is_available():
+            devices.append("mps")
+        devices.append("cpu")
+
+        self.device_opts = devices
+        self.device = tk.StringVar(root, devices[0])

ui/tabs/file_selector.py

@@ -0,0 +1 @@
+# new structure coming soon

ui/tabs/main.py

@@ -0,0 +1 @@
+# new structure coming soon

ui/tabs/settings.py

@@ -0,0 +1 @@
+# new structure coming soon

ui/tooltip.py

@@ -0,0 +1,27 @@
+import customtkinter as ctk
+
+
+class ToolTip(ctk.CTkToplevel):
+    def __init__(self, widget, text):
+        super().__init__()
+        self.withdraw()
+        self.overrideredirect(True)
+        self.attributes("-topmost", True)
+
+        self.label = ctk.CTkLabel(
+            self, text=text, fg_color="gray20", corner_radius=6, padx=10, pady=5
+        )
+        self.label.pack()
+
+        self.widget = widget
+        widget.bind("<Enter>", self.show_tooltip)
+        widget.bind("<Leave>", self.hide_tooltip)
+
+    def show_tooltip(self, event=None):
+        x = self.widget.winfo_rootx() + 20
+        y = self.widget.winfo_rooty() + 20
+        self.geometry(f"+{x}+{y}")
+        self.deiconify()
+
+    def hide_tooltip(self, event=None):
+        self.withdraw()

ui/ui.py

@@ -0,0 +1,480 @@
+import os
+import queue
+import sys
+import json
+import threading
+import tkinter as tk
+from tkinter import filedialog, messagebox, scrolledtext
+from PIL import Image, ImageTk
+
+import customtkinter as ctk
+import torch
+
+from transcription.audio_transcription import AudioTranscription
+from transcription.configuration import Configuration
+from transcription.torch_checker import checkTorch
+from ui.tooltip import ToolTip
+from ui.ui_log_handler import setup_ui_logger
+from utils.requests_to_api import LLMrequest
+
+WINDOW_WIDTH = 1000
+WINDOW_HEIGHT = 725
+
+class TranscriberApp(ctk.CTk):
+    def __init__(self):
+        super().__init__()
+        self.title("Notecast")
+        self.geometry(f"{WINDOW_WIDTH}x{WINDOW_HEIGHT}")
+        ctk.set_appearance_mode("System")
+        ctk.set_default_color_theme("blue")
+
+        # TODO: fix this stuff        
+        base_path = getattr(sys, "_MEIPASS", os.path.dirname(os.path.abspath(__file__)))
+        icon_path = os.path.join(base_path, "assets", "logo.ico")
+        self.iconbitmap(icon_path)
+
+        # states
+        self.progress_queue = queue.Queue()
+        self.transcribe_thread = None
+        self.stop_flag = threading.Event()
+
+        # USER VARIABLES
+        # transcription model settings
+        self.model_var = tk.StringVar(value="openai/whisper-large-v3-turbo")
+        self.batch_var = tk.StringVar(value="32")
+        self.chunk_var = tk.StringVar(value="30")
+        self.dtype_var = tk.StringVar(value="torch.float16")
+        self.transcription_lang_var = tk.StringVar(value="ru")
+        # llm settings
+        self.conspect_transcription_lang_var = tk.StringVar(value="Russian")
+        self.api_key_var = tk.StringVar(value="")
+        self.base_url_var = tk.StringVar(value="")
+        self.api_model_var = tk.StringVar(value="")
+        # checkboxes
+        self.create_conspect = tk.BooleanVar(value=False)
+        self.remove_transcription = tk.BooleanVar(value=False)
+        self.latex_compiler = tk.StringVar(value="")
+
+        # settings device options
+        device_opts = []
+        if torch.cuda.is_available():
+            device_opts.append("cuda")
+        if torch.backends.mps.is_available():
+            device_opts.append("mps")
+        device_opts.append("cpu")
+        self.device_var = tk.StringVar(value=device_opts[0])
+        self.device_opts = device_opts
+
+        # input & output file variables
+        self.input_file_var = tk.StringVar()
+        self.output_file_var = tk.StringVar()
+
+        # tabs packing
+        self.tabview = ctk.CTkTabview(self)
+        self.tabview.pack(expand=True, fill="both", padx=10, pady=10)
+
+        self.transcript_tab = self.tabview.add("Transcription")
+        self.settings_tab = self.tabview.add("Settings")
+
+        self._build_transcription_tab()
+        self._build_settings_tab()
+
+        # logger
+        self.ui_logger = setup_ui_logger(self.log_box)
+
+    ### TRANSCRIPTION TAB
+    def _build_transcription_tab(self):
+        # file selectors
+        file_frame = ctk.CTkFrame(self.transcript_tab, corner_radius=10)
+        file_frame.pack(padx=20, pady=10, fill="x")
+
+        ctk.CTkLabel(file_frame, text="Input file:").pack(side="left", padx=5, pady=5)
+        ctk.CTkEntry(file_frame, textvariable=self.input_file_var, width=400).pack(
+            side="left", padx=5, pady=5, expand=True, fill="x"
+        )
+        ctk.CTkButton(file_frame, text="Browse", command=self._browse_input).pack(
+            side="left", padx=5, pady=5
+        )
+
+        out_frame = ctk.CTkFrame(self.transcript_tab, corner_radius=10)
+        out_frame.pack(padx=20, pady=5, fill="x")
+        ctk.CTkLabel(out_frame, text="Output file:").pack(side="left", padx=5, pady=5)
+        ctk.CTkEntry(out_frame, textvariable=self.output_file_var, width=400).pack(
+            side="left", padx=5, pady=5, expand=True, fill="x"
+        )
+        ctk.CTkButton(out_frame, text="Browse", command=self._browse_output).pack(
+            side="left", padx=5, pady=5
+        )
+
+        # controls
+        ctrl_frame = ctk.CTkFrame(self.transcript_tab, corner_radius=10)
+        ctrl_frame.pack(padx=20, pady=10, fill="x")
+
+        ctk.CTkButton(ctrl_frame, text="Check Torch", command=self._check_torch).pack(
+            side="left", padx=10, pady=5
+        )
+        self.start_button = ctk.CTkButton(
+            ctrl_frame, text="Start", command=self._start_transcription
+        )
+        self.start_button.pack(side="left", padx=10, pady=5)
+
+        self.stop_button = ctk.CTkButton(
+            ctrl_frame, text="Stop", command=self._stop_transcription, state="disabled"
+        )
+        self.stop_button.pack(side="left", padx=10, pady=5)
+
+        self.create_conspect_checkbox = ctk.CTkCheckBox(
+            ctrl_frame,
+            text="Create conspect",
+            variable=self.create_conspect,
+            onvalue=True,
+            offvalue=False,
+        )
+        self.create_conspect_checkbox.pack(side="left", padx=10, pady=5)
+        
+        self.remove_transcription_checkbox = ctk.CTkCheckBox(
+            ctrl_frame,
+            text="Remove transcription file after",
+            variable=self.remove_transcription,
+            onvalue=True,
+            offvalue=False,
+        )
+        self.remove_transcription_checkbox.pack(side="left", padx=10, pady=5)
+
+        # TODO: add unload model button here
+
+        # log box
+        self.log_box = scrolledtext.ScrolledText(
+            self.transcript_tab,
+            wrap="word",
+            height=20,
+            font=("Consolas", 16),
+        )
+        self.log_box.pack(padx=20, pady=10, expand=True, fill="both")
+
+    ### SETTINGS TAB
+    def _build_settings_tab(self):
+        pad = 20
+
+        def add_setting(parent, row, col, text, tooltip, variable, values: list | None):
+            frame = ctk.CTkFrame(parent)
+            frame.grid(row=row, column=col, padx=pad, pady=(pad, 5), sticky="nsew")
+
+            label = ctk.CTkLabel(frame, text=text)
+            label.grid(row=0, column=0, sticky="w")
+            help_icon = ctk.CTkLabel(frame, text="?", width=20, cursor="question_arrow")
+            help_icon.grid(row=0, column=1, sticky="w", padx=(5, 0))
+            ToolTip(help_icon, tooltip)
+
+            if values:
+                ctk.CTkOptionMenu(frame, variable=variable, values=values).grid(
+                    row=1, column=0, columnspan=2, sticky="ew", pady=(5, 0)
+                )
+            else:
+                ctk.CTkEntry(frame, textvariable=variable).grid( 
+                    row=1, column=0, columnspan=2, sticky="ew", pady=(5, 0)
+                )
+
+            frame.grid_columnconfigure(0, weight=1)
+
+        grid = ctk.CTkFrame(self.settings_tab)
+        grid.pack(fill="both", expand=True)
+
+        # first row
+        ### Model setting
+        add_setting(
+            parent=grid,
+            row=0,
+            col=0,
+            text="Model:",
+            tooltip="Choose model for speech recognition",
+            variable=self.model_var,
+            values=[
+                "openai/whisper-large-v3-turbo",
+                "openai/whisper-large-v2",
+                "openai/whisper-large",
+                "openai/whisper-medium",
+                "openai/whisper-small",
+                "openai/whisper-tiny",
+            ],
+        )
+        ### Batch size setting
+        add_setting(
+            parent=grid,
+            row=0,
+            col=1,
+            text="Batch size:",
+            tooltip="Chunks count for one iteration",
+            variable=self.batch_var,
+            values=["32", "16", "8", "4", "2"],
+        )
+
+        # second row
+        ### Data type setting
+        add_setting(
+            parent=grid,
+            row=1,
+            col=0,
+            text="Data type:",
+            tooltip="Data type for calculations",
+            variable=self.dtype_var,
+            values=["torch.float16", "torch.float32", "torch.bfloat16"],
+        )
+        ### Chunk Length setting
+        add_setting(
+            parent=grid,
+            row=1,
+            col=1,
+            text="Chunk length (s):",
+            tooltip="Maximum length of processing audio fragment",
+            variable=self.chunk_var,
+            values=["30", "24", "20", "14", "10", "6"],
+        )
+
+        # third row
+        ### Device setting
+        add_setting(
+            parent=grid,
+            row=2,
+            col=0,
+            text="Device:",
+            tooltip="Choose device\n- CUDA for CUDA & ROCm\n- MPS for Apple Silicon  \n- CPU for CPU-only mode",
+            variable=self.device_var,
+            values=self.device_opts,
+        )
+        ### Transcription language setting
+        add_setting(
+            parent=grid,
+            row=2,
+            col=1,
+            text="Transcription language:",
+            tooltip="Choose the transcription language",
+            variable=self.transcription_lang_var,
+            values=["ru", "en"],
+        )
+
+        # fourth row
+        ### OpenAI API key setting
+        add_setting(
+            parent=grid,
+            row=3,
+            col=0,
+            text="Insert OpenAI API key here:",
+            tooltip="Give this programm access to LLM that would create a fully prepared conspect with AI overviews",
+            variable=self.api_key_var,
+            values=None,
+        )
+        ### Model name setting
+        add_setting(
+            parent=grid,
+            row=3,
+            col=1,
+            text="Model name:",
+            tooltip="Name of the model that you are going to use",
+            variable=self.api_model_var,
+            values=None,
+        )
+
+        # fifth row
+        ### Base URL setting
+        add_setting(
+            parent=grid,
+            row=4,
+            col=0,
+            text="Base URL:",
+            tooltip="OpenAI base URL. Blank for None.",
+            variable=self.base_url_var,
+            values=None,
+        )
+        ### Output (conspect) language setting
+        add_setting(
+            parent=grid,
+            row=4,
+            col=1,
+            text="Conspect language:",
+            tooltip="Conspect language. Blank for English (default)",
+            variable=self.conspect_transcription_lang_var,
+            values=None,
+        )
+        add_setting(
+            parent=grid,
+            row=4,
+            col=0,
+            text="LaTeX compiler:",
+            tooltip="Choose LaTeX compiler",
+            variable=self.latex_compiler,
+            values=[
+                "latexmk",
+                "lualatex",
+                "xelatex",
+                "bibtex"
+            ]
+        )
+
+        ### Custom Prompt setting
+        customPromptFrame = ctk.CTkFrame(grid)
+        customPromptFrame.grid(
+            row=5, column=0, columnspan=2, padx=20, pady=(20, 5), sticky="nsew"
+        )
+
+        label = ctk.CTkLabel(customPromptFrame, text="Custom Prompt:")
+        label.grid(row=0, column=0, sticky="w")
+
+        help_icon = ctk.CTkLabel(
+            customPromptFrame, text="?", width=20, cursor="question_arrow"
+        )
+        help_icon.grid(row=0, column=1, sticky="w", padx=(5, 0))
+        ToolTip(
+            help_icon,
+            "Enter your custom prompt for model.",
+        )
+
+        self.custom_prompt_textbox = ctk.CTkTextbox(
+            customPromptFrame, width=400, height=150
+        )
+        self.custom_prompt_textbox.grid(
+            row=1, column=0, columnspan=2, sticky="nsew", pady=(5, 0)
+        )
+
+        customPromptFrame.grid_columnconfigure(0, weight=1)
+        customPromptFrame.grid_rowconfigure(1, weight=1)
+
+        grid.grid_columnconfigure((0, 1), weight=1)
+
+    # action buttons
+    def _browse_input(self):
+        path = filedialog.askopenfilename(
+            title="Select input audio file",
+            filetypes=[
+                ("Media files", "*.wav *.mp3 *.m4a *.flac *.ogg *.mp4 *.mkv *.avi"),
+                ("Audio files", "*.wav *.mp3 *.m4a *.flac *.ogg"),
+                ("Video files", "*.mp4 *.mkv *.avi"),
+                ("All files", "*.*"),
+            ],
+        )
+        if path:
+            self.input_file_var.set(path)
+
+    def _browse_output(self):
+        # TODO: add custom filename here
+        directory = filedialog.askdirectory(
+            title="Select output directory",
+        )
+        if directory:
+            input_name = os.path.basename(self.input_file_var.get())
+            name, _ = os.path.splitext(input_name)
+            # TODO: redo output_name logic maybe?
+            output_name = f"{"".join(name.split(".")[:-1:])}.txt"
+            path = os.path.join(directory, output_name)
+
+            self.output_file_var.set(path)
+
+    def _check_torch(self):
+        checkTorch(self.ui_logger)
+        
+        self.ui_logger.info(f"==== Transcription ====")
+        self.ui_logger.info(f"Transcription model: {self.model_var.get()}")
+        self.ui_logger.info(f"Batch size (in chunks): {self.batch_var.get()}")
+        self.ui_logger.info(f"Chunk size (in seconds): {self.chunk_var.get()}")
+        self.ui_logger.info(f"Data type: {self.dtype_var.get()}")
+        self.ui_logger.info(f"Transcription language: {self.transcription_lang_var.get()}")
+        self.ui_logger.info(f"=======================")
+        
+        debug_api_key_var: str = self.api_key_var.get() if self.api_key_var.get() else "Not set"
+        debug_api_model_var: str = self.api_model_var.get() if self.api_model_var.get() else "Not set"
+        debug_base_url_var: str = self.base_url_var.get() if self.base_url_var.get() else "Not set"
+        debug_transcription_lang_var = self.transcription_lang_var.get() if self.transcription_lang_var.get() else "Not set"
+        debug_custom_prompt_var: str = self.custom_prompt_textbox.get() if self.custom_prompt_textbox.get() else "Not set"
+        
+        self.ui_logger.info(f"===== LLM Setting =====")
+        self.ui_logger.info(f"API key: {debug_api_key_var}")
+        self.ui_logger.info(f"Model name setting: {debug_api_model_var}")
+        self.ui_logger.info(f"Base URL setting: {debug_base_url_var}")
+        # self.ui_logger.info(f"Custom prompt: {debug_transcription_lang_var}") # <-- issue here
+        self.ui_logger.info(f"=======================")
+        
+
+    def _start_transcription(self):
+        infile = self.input_file_var.get().strip()
+        if not infile or not os.path.isfile(infile):
+            messagebox.showerror("Error", "Please select a valid input file.")
+            return
+
+        self.start_button.configure(state="disabled")
+        self.stop_button.configure(state="normal")
+        self.ui_logger.info("Starting transcription...")
+
+        self.stop_flag.clear()
+        self.transcribe_thread = threading.Thread(
+            target=self._transcribe_worker, args=(infile,), daemon=True
+        )
+        self.transcribe_thread.start()
+
+    def _stop_transcription(self, Audio: AudioTranscription):
+        self.stop_flag.set()
+        self.ui_logger.info("Stopping transcription...")
+        self.ui_logger.info("Unloading model...")
+        Audio._unload_model()
+        self.stop_button.configure(state="disabled")
+
+    def _transcribe_worker(self, infile: str):
+        try:
+            config = Configuration(
+                device=self.device_var.get(),
+                modelName=self.model_var.get(),
+                batchSize=int(self.batch_var.get()),
+                chunkSize=int(self.chunk_var.get()),
+                data_type=self.dtype_var.get(),
+            )
+            Audio = AudioTranscription(
+                filepath=infile,
+                config=config,
+                language=self.transcription_lang_var.get(),
+            )
+            transcription = Audio.transcribeAudio()
+            
+            # remove from here
+            outfile = self.output_file_var.get().strip()
+            if not outfile:
+                outfile = infile
+            
+            if not self.remove_transcription.get():
+                outfile += ".txt"
+
+                with open(outfile, "w", encoding="utf-8") as f:
+                    f.write(transcription)
+                self.ui_logger.info(f"Transcription saved to {outfile}.")
+            
+            # if self.create_conspect.get():
+            #     # TODO: add custom prompt ability here
+            #     # TODO: add logging here
+            #     # TODO: add progressbar instead of tqdm
+            #     self.ui_logger.info(f"Starting creating conspect via {self.api_model_var.get()}...")
+            #     with open("utils/prompts/default_prompt.txt", "r", encoding="utf-8") as f:
+            #         default_prompt = "\n".join(f.readlines())
+                
+            #     # if self.custom_prompt_textbox.get(): # <-- issue here
+            #     #     prompt = transcription + "\n" + self.custom_prompt_textbox.get()
+            #     # else:
+            #     #     prompt = transcription + "\n" + default_prompt
+                
+            #     prompt = transcription + "\n" + default_prompt
+
+            #     request = LLMrequest(
+            #         api_key=self.api_key_var.get(),
+            #         model_name=self.api_model_var.get(),
+            #         base_url=self.base_url_var.get(),
+            #     )
+            #     response = request.get_response(prompt=prompt)
+            #     outfile += ".md"
+            #     with open(outfile, "w", encoding="utf-8") as f:
+            #         f.write(response)
+                
+            #     self.ui_logger.info(f"Conspect saved to {outfile}.")
+
+        except Exception as e:
+            self.ui_logger.error(f"Error: {e}")
+            messagebox.showerror("Error", str(e))
+        finally:
+            self.start_button.configure(state="normal")
+            self.stop_button.configure(state="disabled")

ui/ui_log_handler.py

@@ -0,0 +1,39 @@
+import logging
+import tkinter as tk
+
+
+class UILogHandler(logging.Handler):
+    def __init__(self, text_widget: tk.Text):
+        super().__init__()
+        self.text_widget = text_widget
+
+    def emit(self, record):
+        log_entry = self.format(record)
+        self.text_widget.insert(tk.END, log_entry + "\n")
+        self.text_widget.see(tk.END)
+
+
+# TODO: status should be here
+def setup_ui_logger(text_widget: tk.Text, level=logging.INFO):
+    logger = logging.getLogger("UI_LOGGER")
+    logger.setLevel(level)
+    logger.handlers.clear()
+
+    handler = UILogHandler(text_widget)
+    handler.setFormatter(logging.Formatter("%(asctime)s - %(levelname)s - %(message)s"))
+    logger.addHandler(handler)
+
+    return logger
+
+# def setup_ui_logger(
+#     text_widget: tk.Text,
+#     base_logger: logging.Logger,
+#     level=logging.INFO,
+# ):
+#     handler = UILogHandler(text_widget)
+#     handler.setLevel(level)
+#     handler.setFormatter(
+#         logging.Formatter("%(asctime)s - %(levelname)s - %(message)s")
+#     )
+
+#     base_logger.addHandler(handler)

utils/exceptions.py

@@ -0,0 +1,13 @@
+# not sure, but why not?
+
+class AudioPreparationError(Exception):
+    """Base exception for audio preparation"""
+    pass
+
+class ResamplingError(AudioPreparationError):
+    """Raises when resampling fails"""
+    pass
+
+class DownmixingError(AudioPreparationError):
+    """Raises when downmixing to mono fails"""
+    pass

utils/logger.py

@@ -1,36 +0,0 @@
-import logging
-import sys
-
-def setup_logger(
-    name: str = __name__,
-    level: int = logging.INFO,
-    log_to_file: bool = False,
-    filename: str = "app.log"
-) -> logging.Logger:
-    """
-    Logger configuration with output to command line and (optional) to file
-
-    :param name: logger name
-    :param level: logging level (DEBUG, INFO, WARNING, ERROR, CRITICAL)
-    :param log_to_file: logging to filename
-    :param filename: filename for logs
-    :return: logger instance
-    """
-    logger = logging.getLogger(name)
-    logger.setLevel(level)
-
-    formatter = logging.Formatter(
-        fmt="%(asctime)s | %(levelname)-8s | %(name)s | %(message)s",
-        datefmt="%Y-%m-%d %H:%M:%S"
-    )
-
-    console_handler = logging.StreamHandler(sys.stdout)
-    console_handler.setFormatter(formatter)
-    logger.addHandler(console_handler)
-
-    if log_to_file:
-        file_handler = logging.FileHandler(filename, encoding="utf-8")
-        file_handler.setFormatter(formatter)
-        logger.addHandler(file_handler)
-
-    return logger

utils/prompts/default_prompt_md.j2

@@ -0,0 +1,58 @@
+Create a complete, structured **Obsidian Markdown** educational summary based on the transcript.
+
+The final document must be written in **{{ language }}**.
+
+Use the transcript strictly as the factual basis:
+- Do NOT invent facts.
+- You may add clarifications and explanations, but they MUST be placed in special blocks.
+- Preserve the logical order and meaning of the original speaker.
+- Transform speech into clear academic exposition.
+- Quotes must be accurate and **<= 25 words**.
+- Do not reference the transcript or the speaker. Write as a stand-alone educational text.
+
+You must follow the template below exactly.
+
+---
+
+# {{ title }}
+
+## Key Ideas
+- bullet points summarizing the main concepts
+
+## Core Concepts and Definitions
+- **Term** — definition
+- **Another Term** — definition
+
+## Detailed Summary
+Write a logically structured, academically clear exposition.  
+Break it into paragraphs, subsections, lists, and quotes.
+
+To include quotes from the transcript, use:
+
+> "{{ quote }}"  
+(Ensure that every quote is ≤ 25 words.)
+
+## Examples and Explanations
+Use examples that appear in the transcript.  
+Additional clarifications must use the following block:
+
+> AI clarification:  
+> Your extended explanation, analogy, or contextual detail.
+
+Short notes should use:
+
+**AI note:** short clarification
+
+## Conclusions and Key Insights
+- summary points
+- what the reader should remember
+
+## Self-Assessment Questions
+- question 1
+- question 2
+- question 3
+
+---
+
+### Transcript (for your reference only):
+{{ transcription }}

utils/prompts/default_prompt_tex.j2

@@ -0,0 +1,75 @@
+Create a clean, structured **LaTeX educational summary** based strictly on the transcript.
+
+The final document must be written in **{{ language }}**.
+
+Guidelines:
+- Use ONLY facts from the transcript as the base.
+- You MAY add explanations, but they MUST appear in designated blocks.
+- Do NOT reference the transcript or the speaker.
+- Maintain the speaker`s logical order.
+- Produce academically clear, structured LaTeX text.
+- Quotes must be accurate and <= 25 words.
+
+Use the structure below.  
+Assume that stylistic environments are already defined in the `.sty` file.
+
+---
+
+\section*{ {{ title }} }
+
+\subsection*{Key Ideas}
+\begin{itemize}
+\item main idea
+\item …
+\end{itemize}
+
+\subsection*{Core Concepts and Definitions}
+\begin{description}
+\item[\textbf{Term}] definition
+\item[\textbf{Another Term}] definition
+\end{description}
+
+\subsection*{Detailed Summary}
+Write a clear, logically structured academic explanation.  
+Break content into paragraphs, subsections, lists.
+
+Quotes from the transcript must be included as:
+
+\begin{quote}
+"{{ quote }}"
+\end{quote}
+
+(Ensure each quote is ≤ 25 words.)
+
+\subsection*{Examples and Explanations}
+
+Examples based on the transcript.
+
+Additional clarifications must use the following custom environment (assumed to exist in the .sty):
+
+\begin{aiclarification}
+Your additional explanation, analogy, or contextual expansion.
+\end{aiclarification}
+
+Short notes should use:
+
+\ainote{short clarification}
+
+\subsection*{Conclusions and Key Insights}
+\begin{itemize}
+\item conclusion
+\item conclusion
+\end{itemize}
+
+\subsection*{Self-Assessment Questions}
+\begin{enumerate}
+\item question
+\item question
+\item question
+\end{enumerate}
+
+---
+
+% Transcript included only for reference (model must not copy or analyze or mention this directly)
+% ======================================================
+% {{ transcription | replace("%", "\\%") }}

utils/prompts/prompt_manager.py

@@ -0,0 +1,78 @@
+from __future__ import annotations
+
+import os
+from jinja2 import Environment, FileSystemLoader, StrictUndefined
+
+
+class PromptManager:
+    """
+    Self-written class for Jinja2 templates especially for summarization.
+
+    Supports:
+    - md
+    - latex
+
+    Usage examples:
+    ```
+    from prompts.prompt_manager import PromptManager
+
+    pm = PromptManager("notecast/prompts")
+
+    markdown_text = pm.render(
+        "markdown_default_prompt.md.j2",
+        language="Russian",
+        title="Calculus - Lecture 1",
+        transcription=raw_transcription_text
+    )
+
+    latex_text = pm.render(
+        "latex_default_prompt.j2",
+        language="Russian",
+        title="Calculus - Lecture 1",
+        transcription=raw_transcription_text
+    )
+    ```
+    """
+
+    def __init__(self, templates_dir: str) -> None:
+        if not os.path.isdir(templates_dir):
+            raise NotADirectoryError(f"Templates directory not found: {templates_dir}")
+
+        self.templates_dir = templates_dir
+
+        self.env = Environment(
+            loader=FileSystemLoader(self.templates_dir),
+            autoescape=False,
+            trim_blocks=True,
+            lstrip_blocks=True,
+            undefined=StrictUndefined
+        )
+
+    def list_templates(self) -> list[str]:
+        templates = []
+        for file in os.listdir(self.templates_dir):
+            if file.endswith(".j2"):
+                templates.append(file)
+        return templates
+
+    def load(self, template_name: str):
+        try:
+            return self.env.get_template(template_name)
+        except Exception as e:
+            raise RuntimeError(f"Failed to load template '{template_name}': {e}")
+
+    def render(
+        self,
+        template_name: str,
+        *,
+        language: str,
+        title: str,
+        transcription: str
+    ) -> str:
+        template = self.load(template_name)
+
+        return template.render(
+            language=language,
+            title=title,
+            transcription=transcription
+        )

utils/requests_to_api.py

@@ -0,0 +1,17 @@
+import openai
+
+# maybe make it asynchronous?
+class LLMrequest:
+    def __init__(self, api_key: str, model_name: str, base_url: str = None):
+        if base_url:
+            self.client = openai.OpenAI(api_key=api_key, base_url=base_url)
+        else:
+            self.client = openai.OpenAI(api_key=api_key)
+        self.model = model_name
+
+    def get_response(self, prompt: str) -> str:
+        response = self.client.chat.completions.create(
+            model=self.model,
+            messages=[{"role": "user", "content": prompt}],
+        )
+        return response.choices[0].message.content

utils/save_to_file.py

@@ -0,0 +1,9 @@
+def save_to_file(
+    thing: str | dict,
+    path: str,
+):
+    if isinstance(thing, dict):
+        pass
+    elif isinstance(thing, str):
+        with open(path, "w") as outputfile:
+            outputfile.write(thing)