change emoji

.pre-commit-config.yaml

@@ -49,4 +49,3 @@ repos:
       - id: poetry-export
         name: poetry export for base requirements
         args: [-f, requirements.txt, -o, requirements.txt, -n, --only=main, --without-hashes]
-        stages: [manual]

src/app.py

@@ -1,21 +1,21 @@
 """Template Demo for IBM Granite Hugging Face spaces."""
 
-import os
 from collections.abc import Iterator
 from datetime import datetime
 from pathlib import Path
 from threading import Thread
 
 import gradio as gr
-import langid
 import spaces
 import torch
 import torchaudio
+from transformers import AutoProcessor, AutoModelForSpeechSeq2Seq, TextIteratorStreamer
+import langid
 from punctuators.models import PunctCapSegModelONNX
-from transformers import AutoModelForSpeechSeq2Seq, AutoProcessor, TextIteratorStreamer
-
 pc_model = PunctCapSegModelONNX.from_pretrained("pcs_en")
 
+from themes.research_monochrome import theme
+
 today_date = datetime.today().strftime("%B %-d, %Y")  # noqa: DTZ002
 
 MODEL_ID = "ibm-granite/granite-speech-3.3-2b"
@@ -31,36 +31,18 @@ model = AutoModelForSpeechSeq2Seq.from_pretrained(
     MODEL_ID, device_map="auto", torch_dtype=torch.bfloat16, offload_folder="offload/"
 )
 
-
-def delete_file(path: str) -> None:
-    """Delete a file if it exists.
-
-    Args:
-        path (str): Path to the file to delete.
-
-    Returns:
-        None
-    """
-    if path and os.path.exists(path):
-        try:
-            os.remove(path)
-            print(f"Deleted old audio file: {path}")
-        except Exception as e:
-            print(f"Warning: could not delete {path}: {e}")
-
-
 @spaces.GPU
-def transcribe(audio_file: str, user_prompt: str, prev_file: str) -> Iterator[str]:
-    """Transcribe function for ASR demo.
+def transcribe(audio_file: str, user_prompt: str) -> Iterator[str]:
+    """transcribe function for ASR demo.
 
     Args:
         audio_file (str): Name of audio file from the user.
         user_prompt (str): Instruction from the user (transcription or translation).
-        prev_file (str): Previously uploaded audio file.
 
     Returns:
         str: The generated transcription/translation of the audio file.
     """
+
     # load wav file
     wav, sr = torchaudio.load(audio_file, normalize=True)
     if wav.shape[0] != 1 or sr != 16000:
@@ -68,40 +50,42 @@ def transcribe(audio_file: str, user_prompt: str, prev_file: str) -> Iterator[st
         wav = torch.mean(wav, dim=0, keepdim=True)  # mono
         wav = torchaudio.functional.resample(wav, sr, 16000)
         sr = 16000
-
-    # SAFE POINT: new audio is good → delete old audio if different
-    if prev_file != "" and prev_file != audio_file:
-        delete_file(prev_file)
-
-    # Update prev_file to the *current* file
-    prev_file = audio_file
-
+    
     # Build messages
     chat = [
-        {"role": "system", "content": SYS_PROMPT},
-        {"role": "user", "content": f"<|audio|>{user_prompt}"},
+        dict(role="system", content=SYS_PROMPT),
+        dict(role="user", content=f"<|audio|>{user_prompt}"),
     ]
-    prompt = tokenizer.apply_chat_template(chat, tokenize=False, add_generation_prompt=True)
+    prompt = tokenizer.apply_chat_template(
+        chat, tokenize=False, add_generation_prompt=True)
 
     # run model
-    model_inputs = processor(prompt, wav, device=model.device, return_tensors="pt").to(model.device)
+    model_inputs = processor(
+        prompt,
+        wav,
+        device=model.device,
+        return_tensors="pt").to(model.device)
     streamer = TextIteratorStreamer(tokenizer, timeout=30.0, skip_prompt=True, skip_special_tokens=True)
-
-    kwargs = dict(**model_inputs, streamer=streamer, max_new_tokens=512, do_sample=False, num_beams=1)
+    
+    kwargs = dict(
+        **model_inputs,
+        streamer=streamer,
+        max_new_tokens=512,
+        do_sample=False,
+        num_beams=1
+    )
     t = Thread(target=model.generate, kwargs=kwargs)
     t.start()
 
     text = ""
     for chunk in streamer:
         text += chunk
-        yield text, prev_file
+        yield text
 
     # Apply cap+punct for English-only
-    if langid.classify(text)[0] == "en":
+    if langid.classify(text)[0] == 'en':
         text = pc_model.infer([text])
-        text = " ".join(text[0]).replace("<unk>", " ").replace("<Unk>", " ")  # map <unk> to space
-        yield text, prev_file
-
+        yield " ".join(text[0])
 
 css_file_path = Path(Path(__file__).parent / "app.css")
 head_file_path = Path(Path(__file__).parent / "app_head.html")
@@ -109,25 +93,25 @@ head_file_path = Path(Path(__file__).parent / "app_head.html")
 with gr.Blocks(fill_height=True, css_paths=css_file_path, head_paths=head_file_path, title=TITLE) as demo:
     gr.Markdown(f"# {TITLE}")
     gr.Markdown(DESCRIPTION)
-
-    # State to store the previously uploaded audio file
-    prev_audio = gr.State(value="")
-
     with gr.Row():
-        audio_input = gr.Audio(type="filepath", label="Upload Audio (16kHz mono preferred)")
+        audio_input = gr.Audio(type="filepath",
+                               label="Upload Audio (16kHz mono preferred)")
         with gr.Column():
             output_text = gr.Textbox(label="Transcription", lines=5)
-            choices = [
+            choices = [ 
                 "Transcribe the speech to text",
-                "Translate the speech to French",
-                "Translate the speech to German",
+                "Translate the speech to French", 
+                "Translate the speech to German", 
                 "Translate the speech to Spanish",
-                "Translate the speech to Portuguese",
+                "Translate the speech to Portuguese" 
             ]
-            user_prompt = gr.Dropdown(
-                label="Prompt", choices=choices, interactive=True, allow_custom_value=True, value=choices[0]
-            )
-        audio_input.play(transcribe, inputs=[audio_input, user_prompt, prev_audio], outputs=[output_text, prev_audio])
+            user_prompt = gr.Dropdown(label="Prompt", choices=choices, interactive=True, allow_custom_value=True, value=choices[0])
+        audio_input.play(
+            transcribe,
+            inputs=[
+                audio_input,
+                user_prompt],
+            outputs=output_text)
 
 if __name__ == "__main__":
     demo.launch()