asr

.gitignore

@@ -0,0 +1 @@
+venv

app.py

@@ -1,89 +1,19 @@
-import torch
-from transformers import AutoTokenizer, AutoModelForCausalLM, AutoModel
-from sentence_transformers import SentenceTransformer
-import PyPDF2
-import faiss
-import numpy as np
-import os
+# app.py
+import gradio as gr
+from asr import transcribe_audio
+
+def process_audio(audio):
+    if audio is None:
+        return "No audio received!"
+    return transcribe_audio(audio)
+
+ui = gr.Interface(
+    fn=process_audio,
+    inputs=gr.Audio(source="microphone", type="filepath"),
+    outputs="text",
+    title="🎤 Whisper ASR Tester",
+    description="Speak into the mic and see the transcribed text using Whisper-tiny."
+)
 
-def extract_text_from_pdf(pdf_path):
-    text = ""
-    if pdf_path.endswith('.txt'):
-        with open(pdf_path, 'r', encoding='utf-8') as f:
-            text = f.read()
-    else:
-        with open(pdf_path, 'rb') as f:
-            reader = PyPDF2.PdfReader(f)
-            for page in reader.pages:
-                text += page.extract_text() or ""
-    return text
-
-def split_text(text, chunk_size=512, overlap=64):
-    words = text.split()
-    chunks = []
-    for i in range(0, len(words), chunk_size - overlap):
-        chunk = " ".join(words[i:i+chunk_size])
-        chunks.append(chunk)
-    return chunks
-
-def build_faiss_index(embedding_model, chunks):
-    embeddings = embedding_model.encode(chunks)
-    index = faiss.IndexFlatL2(embeddings.shape[1])
-    index.add(np.array(embeddings))
-    return index, embeddings
-
-def get_top_k_chunks(query, chunks, embedding_model, index, k=5):
-    query_vec = embedding_model.encode([query])
-    D, I = index.search(np.array(query_vec), k)
-    return [chunks[i] for i in I[0]]
-
-def setup_models():
-    model_name = "TinyLlama/TinyLlama-1.1B-Chat-v1.0"
-    tokenizer = AutoTokenizer.from_pretrained(model_name)
-    model = AutoModelForCausalLM.from_pretrained(model_name)
-    embedding_model = SentenceTransformer("sentence-transformers/all-MiniLM-L6-v2")
-    return tokenizer, model, embedding_model
-
-def generate_response(tokenizer, model, context_chunks, query):
-    context = "\n".join(context_chunks)
-    prompt = f"""<|system|>
-You are a helpful assistant. Use the context below to answer the user's question.
-
-CONTEXT:
-{context}
-
-<|user|>
-{query}
-
-<|assistant|>"""
-
-    inputs = tokenizer(prompt, return_tensors="pt", truncation=True, max_length=1024)
-
-    with torch.no_grad():
-        outputs = model.generate(
-            inputs.input_ids,
-            max_length=2048,
-            temperature=0.7,
-            do_sample=True,
-            pad_token_id=tokenizer.eos_token_id,
-        )
-
-    response = tokenizer.decode(outputs[0][inputs['input_ids'].shape[1]:], skip_special_tokens=True)
-    return response.strip()
-
-# =====================
 if __name__ == "__main__":
-    pdf_path = "./doc.txt"  # Or .pdf
-    query = "What is the main topic of the document?"
-
-    # Setup
-    text = extract_text_from_pdf(pdf_path)
-    chunks = split_text(text)
-    tokenizer, model, embedding_model = setup_models()
-    index, _ = build_faiss_index(embedding_model, chunks)
-
-    # Retrieval + Generation
-    top_chunks = get_top_k_chunks(query, chunks, embedding_model, index)
-    response = generate_response(tokenizer, model, top_chunks, query)
-
-    print("Response:\n", response)
+    ui.launch()

asr.py

@@ -0,0 +1,8 @@
+# asr.py
+import whisper
+
+model = whisper.load_model("tiny")  # lightweight + fast
+
+def transcribe_audio(file_path: str) -> str:
+    result = model.transcribe(file_path)
+    return result["text"]

requirements.txt

@@ -1,6 +1,6 @@
-torch
+gradio
+openai-whisper
 transformers
-datasets
-sentence-transformers 
-faiss-cpu 
-PyPDF2
+torch
+TTS
+accelerate