Spaces:
Runtime error
Runtime error
Sync from GitHub via hub-sync
Browse files- Dockerfile +20 -0
- README.md +4 -5
- app.py +8 -0
- config/constants.py +2 -0
- config/settings.py +19 -0
- index.py +63 -0
- models/__init__.py +0 -0
- requirements.txt +63 -0
- utils/__init__.py +0 -0
- utils/audio_cleaning.py +29 -0
- utils/debug_utils.py +26 -0
Dockerfile
ADDED
|
@@ -0,0 +1,20 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
FROM python:3.10-slim
|
| 2 |
+
|
| 3 |
+
RUN apt-get update && apt-get install -y ffmpeg && rm -rf /var/lib/apt/lists/*
|
| 4 |
+
|
| 5 |
+
# Create User
|
| 6 |
+
RUN useradd -m -u 1000 user
|
| 7 |
+
USER user
|
| 8 |
+
ENV PATH="/home/user/.local/bin:$PATH"
|
| 9 |
+
|
| 10 |
+
WORKDIR /app
|
| 11 |
+
|
| 12 |
+
# Install requirements
|
| 13 |
+
COPY --chown=user requirements.txt .
|
| 14 |
+
RUN pip install --no-cache-dir -r requirements.txt
|
| 15 |
+
|
| 16 |
+
# Copy code
|
| 17 |
+
COPY --chown=user . .
|
| 18 |
+
|
| 19 |
+
# HF requires port 7860
|
| 20 |
+
CMD ["uvicorn", "app:app", "--host", "0.0.0.0", "--port", "7860"]
|
README.md
CHANGED
|
@@ -1,12 +1,11 @@
|
|
| 1 |
---
|
| 2 |
-
title:
|
| 3 |
-
emoji:
|
| 4 |
colorFrom: blue
|
| 5 |
colorTo: pink
|
| 6 |
sdk: docker
|
|
|
|
| 7 |
pinned: false
|
| 8 |
-
license: cc
|
| 9 |
-
short_description: backend to transcribe audio input the user gives
|
| 10 |
---
|
| 11 |
|
| 12 |
-
|
|
|
|
| 1 |
---
|
| 2 |
+
title: MedScribe-AI
|
| 3 |
+
emoji: 🩻
|
| 4 |
colorFrom: blue
|
| 5 |
colorTo: pink
|
| 6 |
sdk: docker
|
| 7 |
+
app_file: app.py
|
| 8 |
pinned: false
|
|
|
|
|
|
|
| 9 |
---
|
| 10 |
|
| 11 |
+
# My Project Title
|
app.py
ADDED
|
@@ -0,0 +1,8 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
from fastapi import FastAPI
|
| 2 |
+
|
| 3 |
+
app = FastAPI()
|
| 4 |
+
|
| 5 |
+
@app.get("/")
|
| 6 |
+
async def root():
|
| 7 |
+
return {"message": "Hello World"}
|
| 8 |
+
|
config/constants.py
ADDED
|
@@ -0,0 +1,2 @@
|
|
|
|
|
|
|
|
|
|
| 1 |
+
GROQ_MODEL_NAME = 'whisper-large-v3-turbo'
|
| 2 |
+
GROQ_TARGET_SR = 16000
|
config/settings.py
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
from functools import lru_cache
|
| 2 |
+
from pydantic_settings import BaseSettings, SettingsConfigDict
|
| 3 |
+
|
| 4 |
+
|
| 5 |
+
class Settings(BaseSettings):
|
| 6 |
+
# Pydantic automatically finds CLERK_SECRET_KEY in your .env.local
|
| 7 |
+
clerk_secret_key: str
|
| 8 |
+
gemini_api_key: str
|
| 9 |
+
groq_secret_key: str
|
| 10 |
+
next_public_clerk_publishable_key: str
|
| 11 |
+
smtp_password: str
|
| 12 |
+
smtp_user: str
|
| 13 |
+
vercel_oidc_token: str
|
| 14 |
+
|
| 15 |
+
model_config = SettingsConfigDict(env_file=".env.local", env_file_encoding="utf-8")
|
| 16 |
+
|
| 17 |
+
@lru_cache
|
| 18 |
+
def get_settings() -> Settings:
|
| 19 |
+
return Settings() # type: ignore
|
index.py
ADDED
|
@@ -0,0 +1,63 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import io
|
| 2 |
+
import numpy as np
|
| 3 |
+
from scipy.io import wavfile
|
| 4 |
+
import librosa
|
| 5 |
+
from fastapi import Depends, FastAPI, File, UploadFile, HTTPException
|
| 6 |
+
from api.utils.audio_cleaning import denoise_audio, normalize_audio
|
| 7 |
+
from api.config.settings import Settings, get_settings
|
| 8 |
+
from api.config import constants
|
| 9 |
+
from groq import Groq
|
| 10 |
+
|
| 11 |
+
groq_client = Groq(api_key=get_settings().groq_secret_key)
|
| 12 |
+
app = FastAPI()
|
| 13 |
+
|
| 14 |
+
|
| 15 |
+
@app.get("/")
|
| 16 |
+
def root():
|
| 17 |
+
return {"status": "ok"}
|
| 18 |
+
|
| 19 |
+
|
| 20 |
+
@app.post("/api/transcribe")
|
| 21 |
+
async def process_and_transcribe(file: UploadFile = File(...)):
|
| 22 |
+
audio_bytes = await file.read()
|
| 23 |
+
print(f"File size: {len(audio_bytes)} bytes")
|
| 24 |
+
if len(audio_bytes) == 0:
|
| 25 |
+
raise HTTPException(400, "Empty file")
|
| 26 |
+
|
| 27 |
+
buffer = io.BytesIO(audio_bytes)
|
| 28 |
+
waveform, sr = librosa.load(buffer, sr=None)
|
| 29 |
+
|
| 30 |
+
cleaned_audio = denoise_audio(waveform, sr)
|
| 31 |
+
cleaned_audio = normalize_audio(cleaned_audio)
|
| 32 |
+
|
| 33 |
+
# prepare audio to be sent
|
| 34 |
+
audio_np = cleaned_audio.detach().cpu().numpy().squeeze()
|
| 35 |
+
audio_np = librosa.resample(audio_np, orig_sr=sr, target_sr=constants.GROQ_TARGET_SR)
|
| 36 |
+
audio_int16 = (audio_np * 32767).astype(np.int16)
|
| 37 |
+
|
| 38 |
+
export_buffer = io.BytesIO()
|
| 39 |
+
wavfile.write(export_buffer, constants.GROQ_TARGET_SR, audio_int16)
|
| 40 |
+
export_buffer.seek(0)
|
| 41 |
+
|
| 42 |
+
try:
|
| 43 |
+
transcription = groq_client.audio.transcriptions.create(
|
| 44 |
+
file=(file.filename, export_buffer.read()),
|
| 45 |
+
model=constants.GROQ_MODEL_NAME,
|
| 46 |
+
response_format="json",
|
| 47 |
+
language="en",
|
| 48 |
+
)
|
| 49 |
+
return {
|
| 50 |
+
"transcript": transcription.text,
|
| 51 |
+
"filename": file.filename,
|
| 52 |
+
"duration_seconds": round(len(waveform) / sr, 2),
|
| 53 |
+
}
|
| 54 |
+
|
| 55 |
+
except Exception as e:
|
| 56 |
+
print(f"Groq API Error: {e}")
|
| 57 |
+
raise HTTPException(500, f"Transcription failed: {str(e)}")
|
| 58 |
+
|
| 59 |
+
|
| 60 |
+
if __name__ == "__main__":
|
| 61 |
+
import uvicorn
|
| 62 |
+
|
| 63 |
+
uvicorn.run("index:app", host="127.0.0.1", port=8000, reload=True)
|
models/__init__.py
ADDED
|
File without changes
|
requirements.txt
ADDED
|
@@ -0,0 +1,63 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# annotated-doc==0.0.4
|
| 2 |
+
# annotated-types==0.7.0
|
| 3 |
+
# anyio==4.13.0
|
| 4 |
+
# appdirs==1.4.4
|
| 5 |
+
# audioread==3.1.0
|
| 6 |
+
# certifi==2026.4.22
|
| 7 |
+
# cffi==2.0.0
|
| 8 |
+
# charset-normalizer==3.4.7
|
| 9 |
+
# click==8.3.3
|
| 10 |
+
# DeepFilterLib==0.5.6
|
| 11 |
+
# DeepFilterNet==0.5.6
|
| 12 |
+
# distro==1.9.0
|
| 13 |
+
# fastapi==0.136.1
|
| 14 |
+
# filelock==3.29.0
|
| 15 |
+
# fsspec==2026.4.0
|
| 16 |
+
# git-filter-repo==2.47.0
|
| 17 |
+
# groq==1.2.0
|
| 18 |
+
# h11==0.16.0
|
| 19 |
+
# httpcore==1.0.9
|
| 20 |
+
# httpx==0.28.1
|
| 21 |
+
# idna==3.15
|
| 22 |
+
# intel-openmp==2021.4.0
|
| 23 |
+
# Jinja2==3.1.6
|
| 24 |
+
# joblib==1.5.3
|
| 25 |
+
# lazy-loader==0.5
|
| 26 |
+
# librosa==0.11.0
|
| 27 |
+
# llvmlite==0.47.0
|
| 28 |
+
# loguru==0.7.3
|
| 29 |
+
# MarkupSafe==3.0.3
|
| 30 |
+
# mkl==2021.4.0
|
| 31 |
+
# mpmath==1.3.0
|
| 32 |
+
# msgpack==1.1.2
|
| 33 |
+
# networkx==3.4.2
|
| 34 |
+
# numba==0.65.1
|
| 35 |
+
# numpy==1.26.4
|
| 36 |
+
# packaging==23.2
|
| 37 |
+
# pillow==12.2.0
|
| 38 |
+
# pooch==1.9.0
|
| 39 |
+
# pycparser==3.0
|
| 40 |
+
# pydantic==2.13.4
|
| 41 |
+
# pydantic-settings==2.14.1
|
| 42 |
+
# pydantic_core==2.46.4
|
| 43 |
+
# python-dotenv==1.2.2
|
| 44 |
+
# python-multipart==0.0.28
|
| 45 |
+
# requests==2.34.2
|
| 46 |
+
# scikit-learn==1.7.2
|
| 47 |
+
# scipy==1.15.3
|
| 48 |
+
# sniffio==1.3.1
|
| 49 |
+
# soundfile==0.13.1
|
| 50 |
+
# soxr==1.1.0
|
| 51 |
+
# starlette==1.0.0
|
| 52 |
+
# sympy==1.14.0
|
| 53 |
+
# tbb==2021.13.1
|
| 54 |
+
# threadpoolctl==3.6.0
|
| 55 |
+
# torch==2.3.0
|
| 56 |
+
# torchaudio==2.3.0
|
| 57 |
+
# torchvision==0.18.0
|
| 58 |
+
# typing-inspection==0.4.2
|
| 59 |
+
# urllib3==2.7.0
|
| 60 |
+
# uvicorn==0.47.0
|
| 61 |
+
|
| 62 |
+
uvicorn
|
| 63 |
+
fastapi
|
utils/__init__.py
ADDED
|
File without changes
|
utils/audio_cleaning.py
ADDED
|
@@ -0,0 +1,29 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import torch
|
| 2 |
+
from df.enhance import init_df, enhance
|
| 3 |
+
import librosa
|
| 4 |
+
|
| 5 |
+
def denoise_audio(waveform, sr):
|
| 6 |
+
model, df_state, _ = init_df()
|
| 7 |
+
|
| 8 |
+
if sr != df_state.sr():
|
| 9 |
+
waveform = librosa.resample(waveform, orig_sr=sr, target_sr=df_state.sr())
|
| 10 |
+
|
| 11 |
+
noisy_audio = torch.from_numpy(waveform).float().unsqueeze(0)
|
| 12 |
+
enhanced_audio = enhance(model, df_state, noisy_audio)
|
| 13 |
+
|
| 14 |
+
return enhanced_audio
|
| 15 |
+
|
| 16 |
+
|
| 17 |
+
def normalize_audio(audio_tensor, target_rms_db=-18.0, max_peak_db=-1.5):
|
| 18 |
+
# 1. Normalize Average RMS energy
|
| 19 |
+
rms = torch.sqrt(torch.mean(audio_tensor**2))
|
| 20 |
+
target_rms = 10 ** (target_rms_db / 20.0)
|
| 21 |
+
audio_tensor = audio_tensor * (target_rms / (rms + 1e-8))
|
| 22 |
+
|
| 23 |
+
# 2. Prevent dynamic clipping via Peak normalization fallback
|
| 24 |
+
max_peak = torch.max(torch.abs(audio_tensor))
|
| 25 |
+
target_peak = 10 ** (max_peak_db / 20.0)
|
| 26 |
+
if max_peak > target_peak:
|
| 27 |
+
audio_tensor = audio_tensor * (target_peak / max_peak)
|
| 28 |
+
|
| 29 |
+
return torch.clamp(audio_tensor, -1.0, 1.0)
|
utils/debug_utils.py
ADDED
|
@@ -0,0 +1,26 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import soundfile as sf
|
| 2 |
+
import torch
|
| 3 |
+
import io
|
| 4 |
+
|
| 5 |
+
|
| 6 |
+
def debug_save(data, sr, label):
|
| 7 |
+
"""
|
| 8 |
+
Saves various audio formats to local files for inspection.
|
| 9 |
+
"""
|
| 10 |
+
filename = f"debug_{label}.wav"
|
| 11 |
+
|
| 12 |
+
# Handle Torch Tensors (Cleaned/Normalized)
|
| 13 |
+
if isinstance(data, torch.Tensor):
|
| 14 |
+
data = data.detach().cpu().numpy().squeeze()
|
| 15 |
+
|
| 16 |
+
# Handle BytesIO (The final buffer)
|
| 17 |
+
if isinstance(data, io.BytesIO):
|
| 18 |
+
with open(filename, "wb") as f:
|
| 19 |
+
f.write(data.getvalue())
|
| 20 |
+
print(f"Saved Buffer to {filename}")
|
| 21 |
+
return
|
| 22 |
+
|
| 23 |
+
# Handle Numpy (Waveform or Int16)
|
| 24 |
+
# Note: soundfile handles both float32 and int16 automatically
|
| 25 |
+
sf.write(filename, data, sr)
|
| 26 |
+
print(f"Saved {label} to {filename}")
|