app.py

import torch
from transformers import AutoModelForSpeechSeq2Seq, AutoProcessor, pipeline
import gradio as gr
import os
import time
import tempfile
from scipy.io.wavfile import write
from pydub import AudioSegment

# Configuration
MODEL_ID = "Crystalcareai/Whisper-Medicalv1"
CHUNK_LENGTH_S = 30
RETURN_TIMESTAMPS = True
device = "cpu"
torch_dtype = torch.float32

# Load model and processor once
print("Loading model...")
model = AutoModelForSpeechSeq2Seq.from_pretrained(
    MODEL_ID,
    torch_dtype=torch_dtype,
    use_safetensors=True
)

model.to(device)
processor = AutoProcessor.from_pretrained(MODEL_ID)

# Build pipeline
pipe = pipeline(
    "automatic-speech-recognition",
    model=model,
    tokenizer=processor.tokenizer,
    feature_extractor=processor.feature_extractor,
    torch_dtype=torch_dtype,
    device=device,
    chunk_length_s=CHUNK_LENGTH_S,
    return_timestamps=RETURN_TIMESTAMPS
)

# split func
def split_audio(path, chunk_len_ms=30_000):
    audio = AudioSegment.from_file(path)
    return [audio[i:i+chunk_len_ms] for i in range(0, len(audio), chunk_len_ms)]

#transcribe chunkwise
def transcribe_chunkwise(audio, transcript_state):

    sr, audio_data = audio

    # Save whole file first
    with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as tmp:
        write(tmp.name, sr, audio_data)
        path = tmp.name
   
    # Split audio into 30s chunks
    audio_chunks = split_audio(path)
    new_entries = []

    for i, chunk in enumerate(audio_chunks):
        with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as ctmp:
            chunk.export(ctmp.name, format="wav")
            text = pipe(ctmp.name)["text"]
            transcript_state.append(text)
        yield "\n".join(transcript_state), transcript_state

#trasncribe func
def transcribe(audio):

    if audio is None:
        return "No audio input detected"

    sr, audio_data = audio 

    with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as tmp:
        write(tmp.name, sr, audio_data)
        tmp_path = tmp.name

    text =  pipe(tmp_path)["text"]
    text = text.replace(". ", ".\n").replace("! ", "!\n").replace("? ", "?\n")

    os.remove(tmp_path)
    return text

with gr.Blocks() as app:

    state = gr.State([])

    mic_input = gr.Audio(sources=["microphone"], label="🎙️ speak")
    mic_run_btn = gr.Button("Trasncribe")
    file_input = gr.Audio(sources=["upload"],label="📁 Upload Audio")
    file_run_btn = gr.Button("Trasncribe")
    output_textbox = gr.Textbox(label="Transcription");

    mic_run_btn.click(fn=transcribe_chunkwise, inputs=[mic_input, state], outputs=output_textbox, api_name="transcribe", show_progress=True)
    file_run_btn.click(fn=transcribe_chunkwise, inputs=[file_input, state], outputs=output_textbox, api_name="transcribe", show_progress=True)

app.launch()