Major updates

- UI is now on customtkinter - new environment.yml file - entry point is now in main.py - minor improvements in audio_transcription
2025-09-12 17:43:53 +03:00
parent 8f41105e4b
commit 7cf4c5259b
5 changed files with 260 additions and 315 deletions
@@ -1,195 +1,16 @@
 name: notecast
 channels:
-  - conda-forge
  - defaults
-  - nvidia
-  - pytorch
+  - conda-forge
 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
+  - torchvision
+  - pytorch
+  - torchaudio
+  - transformers
+  - accelerate
+  - ffmpeg
+  - pyinstaller
+  - customtkinter
+  - isort
+  - mypy
+  - black
@@ -1,13 +1,5 @@
-from transcription.audio_transcription import AudioTranscription
-from transcription.torch_checker import check_torch
-from utils.logger import setup_logger
+from ui.ui import TranscriberApp

-logger = setup_logger("main")
-
-check_torch()
-
-try:
-    track = AudioTranscription("sample.mp3")
-    print(track.transcribe_audio())
-except Exception as e:
-    logger.error(f"Execution error: {e}") 
+if __name__ == "__main__":
+    app = TranscriberApp()
+    app.mainloop()
@@ -8,7 +8,7 @@ import time
 import math
 from tqdm import tqdm

-# TODO: rename naming
+# TODO: implement transcription with shift
 class AudioTranscription:
    model_name = "openai/whisper-large-v2" 
    
@@ -53,6 +53,7 @@ class AudioTranscription:
        
        self.chunks: list = []
        self.batches: list = []
+        self.all_transcription: list = []
        try:
            logger.info("Loading model WhisperProcessor...")
            self.processor = WhisperProcessor.from_pretrained(self.model_name)
@@ -71,15 +72,15 @@ class AudioTranscription:
            logger.error(f"Unable to load file {self.filepath}: {e}")
            raise
    
-    def resample(self) -> None:
+    def _resample(self) -> None:
        self.waveform = torchaudio.functional.resample(self.waveform, self.sampling_rate, 16000)
    
-    def to_mono(self):
+    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:
+    def _split_to_chunks(self, chunk_length_s: int = 30) -> None:
        self.logger.info(f"Splitting audio on chunks...")
        
        self.chunk_size = chunk_length_s * 16000  # 16kHz after resampling
@@ -95,15 +96,14 @@ class AudioTranscription:
            chunk = self.waveform[start:end].cpu().numpy().astype("float32")
            self.chunks.append(chunk)
    
-    def resplit_to_batches(self) -> None:
+    def _resplit_to_batches(self) -> None:
        self.logger.info(f"Splitting chunks into batches...")
        self.batches = []
        for i in range(0, len(self.chunks), self.custom_batch_length):
            batch = self.chunks[i:i + self.custom_batch_length]
            self.batches.append(batch)
    
-    
-    def process_all_batches(self) -> None:
+    def _process_all_batches(self) -> None:
        start_time = time.time()
        try:
            self.all_transcription = []
@@ -136,9 +136,10 @@ class AudioTranscription:

    
    def transcribe_audio(self) -> str:
-        self.resample()
-        self.to_mono()
-        self.split_to_chunks()
-        self.resplit_to_batches()
-        self.process_all_batches()
+        # TODO: maybe something else, not str?
+        self._resample()
+        self._to_mono()
+        self._split_to_chunks()
+        self._resplit_to_batches()
+        self._process_all_batches()
        return " ".join(self.all_transcription)
@@ -0,0 +1,9 @@
+from transcription.device_configuration import DeviceConfiguration
+
+# 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
+    }
@@ -1,123 +1,245 @@
+import os
+import threading
+import queue
 import tkinter as tk
-from tkinter.scrolledtext import ScrolledText
-from ui.ui_log_handler import UILogHandler, setup_ui_logger
+from tkinter import scrolledtext, filedialog, messagebox
+
+import customtkinter as ctk
+import torch
+
+from ui.ui_log_handler import setup_ui_logger
 from transcription.torch_checker import check_torch
 from transcription.device_configuration import DeviceConfiguration
 from transcription.audio_transcription import AudioTranscription

-def main():
-    root = tk.Tk()
-    root.title("Audio Transcriptor")
-    root.geometry("800x600")

-    for col in range(4):
-        root.grid_columnconfigure(col, weight=1)
-    root.grid_rowconfigure(6, weight=1)
+WINDOW_WIDTH = 900
+WINDOW_HEIGHT = 650

-    ### Buttons selector
-    check_torch_baton = tk.Button(root, text="Check Torch")
-    check_torch_baton.grid(row=0, column=0, padx=5, pady=5, sticky="ew")

-    # TODO: implement saving function
-    save_configuration_baton = tk.Button(root, text="Save configuration")
-    save_configuration_baton.grid(row=0, column=1, padx=5, pady=5, sticky="ew")
+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: implement deleting function
-    delete_configuration_baton = tk.Button(root, text="Delete configuration")
-    delete_configuration_baton.grid(row=0, column=2, padx=5, pady=5, sticky="ew")
+        # states
+        self.progress_queue = queue.Queue()
+        self.transcribe_thread = None
+        self.stop_flag = threading.Event()

-    start_transcription_baton = tk.Button(root, text="Transcript")
-    start_transcription_baton.grid(row=0, column=3, padx=5, pady=5, sticky="ew")
+        # user variables
+        self.model_var = tk.StringVar(value="openai/whisper-large-v2")
+        self.batch_var = tk.StringVar(value="32")
+        self.dtype_var = tk.StringVar(value="torch.float16")
+        self.chunk_var = tk.StringVar(value="30")

-    ### Model options selector
-    model_options = [
-        "openai/whisper-large-v2",
-        "openai/whisper-large",
-        "openai/whisper-medium",
-        "openai/whisper-small",
-        "openai/whisper-tiny"
-    ]
-    selected_model = tk.StringVar(value=model_options[0])
-    label_model = tk.Label(root, text="Model name:")
-    label_model.grid(row=1, column=0, sticky="w", pady=5, padx=5)
-    dropdown_model_selection = tk.OptionMenu(root, selected_model, *model_options)
-    dropdown_model_selection.grid(row=1, column=1, sticky="ew", pady=5, padx=5)
+        # 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

-    ### Batch size selector
-    batch_sizes = ["32", "16", "8", "4", "2"]
-    selected_batch_size = tk.StringVar(value=batch_sizes[0])
-    label_batch_size = tk.Label(root, text="Batch size:")
-    label_batch_size.grid(row=1, column=2, sticky="w", pady=5, padx=5)
-    dropdown_batch_size_selection = tk.OptionMenu(root, selected_batch_size, *batch_sizes)
-    dropdown_batch_size_selection.grid(row=1, column=3, sticky="ew", pady=5, padx=5)
+        self.input_file_var = tk.StringVar()
+        self.output_file_var = tk.StringVar()

-    ### Data type selector
-    data_types = ["torch.float16", "torch.float32", "torch.bfloat16"]
-    selected_data_type = tk.StringVar(value=data_types[0])
-    label_data_type = tk.Label(root, text="Data type:")
-    label_data_type.grid(row=2, column=0, sticky="w", pady=5, padx=5)
-    dropdown_data_type_selection = tk.OptionMenu(root, selected_data_type, *data_types)
-    dropdown_data_type_selection.grid(row=2, column=1, sticky="ew", pady=5, padx=5)
+        # tabs packing
+        self.tabview = ctk.CTkTabview(self)
+        self.tabview.pack(expand=True, fill="both", padx=10, pady=10)

-    ### Chunk length selector
-    chunk_lengths = ["30", "25", "20", "15", "10", "5"]
-    selected_chunk_length = tk.StringVar(value=chunk_lengths[0])
-    label_chunk_length = tk.Label(root, text="Chunk length:")
-    label_chunk_length.grid(row=2, column=2, sticky="w", pady=5, padx=5)
-    dropdown_chunk_length_selection = tk.OptionMenu(root, selected_chunk_length, *chunk_lengths)
-    dropdown_chunk_length_selection.grid(row=2, column=3, sticky="ew", pady=5, padx=5)
+        self.transcript_tab = self.tabview.add("Transcription")
+        self.settings_tab = self.tabview.add("Settings")

-    # TODO: add device selector (cuda/mps/cpu)
+        self._build_transcription_tab()
+        self._build_settings_tab()

-    ### Filepath (input)
-    # TODO: add checker if path is valid/invalid (i think in utils or something)
-    label_file_path = tk.Label(root, text="Input filepath:")
-    label_file_path.grid(row=3, column=0, sticky="w", pady=5, padx=5)
-    file_path = tk.Text(root, height=1)
-    file_path.grid(row=3, column=1, columnspan=3, sticky="ew", pady=5, padx=5)
+        # logger
+        self.ui_logger = setup_ui_logger(self.log_box)

-    ### Filepath (output)
-    # TODO: add question mark here with tip while mouse is on it
-    label_output_file_path = tk.Label(root, text="Output filepath:")
-    label_output_file_path.grid(row=4, column=0, sticky="w", pady=5, padx=5)
-    output_file_path = tk.Text(root, height=1)
-    output_file_path.grid(row=4, column=1, columnspan=3, sticky="ew", pady=5, padx=5)
+    # main 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")

-    def show_selections():
-        ui_logger.info(f"Selected model: {selected_model.get()}")
-        ui_logger.info(f"Selected batch size: {selected_batch_size.get()} chunks")
-        ui_logger.info(f"Selected data type: {selected_data_type.get()}")
-    
-    show_selections_baton = tk.Button(root, text="Show Selections", command=show_selections)
-    show_selections_baton.grid(row=5, column=0, columnspan=4, pady=5, sticky="ew")
-
-    log_box = ScrolledText(root, wrap="word")
-    log_box.grid(row=6, column=0, columnspan=4, sticky="nsew", padx=10, pady=5)
-    ui_logger = setup_ui_logger(log_box)
-    
-    def transcribe():
-        current_device_config = DeviceConfiguration(
-            device="cuda",
-            model_name=selected_model.get(),
-            batch_size=int(selected_batch_size.get()),
-            chunk_length_s=30,
-            data_type=selected_data_type.get()
+        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"
        )
-        Audio = AudioTranscription(
-            filepath=file_path.get("1.0", "end-1c"),
-            device_configuration=current_device_config,
-            logger=ui_logger
+        ctk.CTkButton(file_frame, text="Browse", command=self._browse_input).pack(
+            side="left", padx=5, pady=5
        )
-        transcription = Audio.transcribe_audio()
-        with open(f"{file_path.get('1.0', 'end-1c')}.txt", "w") as output_file:
-            output_file.write(transcription)

-    check_torch_baton.config(command=lambda: check_torch(ui_logger))
-    start_transcription_baton.config(command=transcribe)
+        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
+        )

-    root.mainloop()
+        # 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_btn = ctk.CTkButton(
+            ctrl_frame, text="Start transcription", command=self._start_transcription
+        )
+        self.start_btn.pack(side="left", padx=10, pady=5)

+        self.stop_btn = ctk.CTkButton(
+            ctrl_frame, text="Stop", command=self._stop_transcription, state="disabled"
+        )
+        self.stop_btn.pack(side="left", padx=10, pady=5)

-if __name__ == "__main__":
-    main()
+        # 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")
+
+    def _build_settings_tab(self):
+        pad = 20
+
+        ### Model
+        ctk.CTkLabel(self.settings_tab, text="Model:").pack(
+            anchor="w", padx=pad, pady=(pad, 5)
+        )
+        ctk.CTkOptionMenu(
+            self.settings_tab,
+            variable=self.model_var,
+            values=[
+                "openai/whisper-large-v2",
+                "openai/whisper-large",
+                "openai/whisper-medium",
+                "openai/whisper-small",
+                "openai/whisper-tiny",
+            ],
+        ).pack(fill="x", padx=pad, pady=5)
+
+        ### Batch size
+        ctk.CTkLabel(self.settings_tab, text="Batch size:").pack(
+            anchor="w", padx=pad, pady=(pad, 5)
+        )
+        ctk.CTkOptionMenu(
+            self.settings_tab,
+            variable=self.batch_var,
+            values=["32", "16", "8", "4", "2"],
+        ).pack(fill="x", padx=pad, pady=5)
+
+        ### Data type
+        ctk.CTkLabel(self.settings_tab, text="Data type:").pack(
+            anchor="w", padx=pad, pady=(pad, 5)
+        )
+        ctk.CTkOptionMenu(
+            self.settings_tab,
+            variable=self.dtype_var,
+            values=["torch.float16", "torch.float32", "torch.bfloat16"],
+        ).pack(fill="x", padx=pad, pady=5)
+
+        ### Chunk length
+        ctk.CTkLabel(self.settings_tab, text="Chunk length (s):").pack(
+            anchor="w", padx=pad, pady=(pad, 5)
+        )
+        ctk.CTkOptionMenu(
+            self.settings_tab,
+            variable=self.chunk_var,
+            values=["30", "25", "20", "15", "10", "5"],
+        ).pack(fill="x", padx=pad, pady=5)
+
+        ### Device
+        ctk.CTkLabel(self.settings_tab, text="Device:").pack(
+            anchor="w", padx=pad, pady=(pad, 5)
+        )
+        ctk.CTkOptionMenu(
+            self.settings_tab,
+            variable=self.device_var,
+            values=self.device_opts,
+        ).pack(fill="x", padx=pad, pady=5)
+
+    # action buttons
+    def _browse_input(self):
+        path = filedialog.askopenfilename(
+            title="Select input audio file",
+            filetypes=[("Audio files", "*.wav *.mp3 *.m4a *.flac"), ("All files", "*.*")]
+        )
+        if path:
+            self.input_file_var.set(path)
+
+    def _browse_output(self):
+        path = filedialog.asksaveasfilename(
+            title="Select output file",
+            defaultextension=".txt",
+            filetypes=[("Text files", "*.txt"), ("All files", "*.*")]
+        )
+        if path:
+            self.output_file_var.set(path)
+
+    def _check_torch(self):
+        check_torch(self.ui_logger)
+
+    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_btn.configure(state="disabled")
+        self.stop_btn.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):
+        self.stop_flag.set()
+        self.ui_logger.info("Stopping transcription...")
+        self.stop_btn.configure(state="disabled")
+
+    def _transcribe_worker(self, infile: str):
+        try:
+            config = DeviceConfiguration(
+                device=self.device_var.get(),
+                model_name=self.model_var.get(),
+                batch_size=int(self.batch_var.get()),
+                chunk_length_s=int(self.chunk_var.get()),
+                data_type=self.dtype_var.get(),
+            )
+            Audio = AudioTranscription(
+                filepath=infile,
+                device_configuration=config,
+                logger=self.ui_logger,
+            )
+            transcription = Audio.transcribe_audio()
+
+            outfile = self.output_file_var.get().strip()
+            if not outfile:
+                outfile = infile + ".txt"
+
+            with open(outfile, "w", encoding="utf-8") as f:
+                f.write(transcription)
+
+            self.ui_logger.info(f"Transcription saved to {outfile}")
+        except Exception as e:
+            self.ui_logger.error(f"Error: {e}")
+            messagebox.showerror("Error", str(e))
+        finally:
+            self.start_btn.configure(state="normal")
+            self.stop_btn.configure(state="disabled")