poc for translation

cuda_env.sh

@@ -0,0 +1,11 @@
+#!/bin/bash
+# Add NVIDIA CUDA libraries to LD_LIBRARY_PATH for faster-whisper
+NVIDIA_LIB_BASE="$(pwd)/.venv/lib/python3.14/site-packages/nvidia"
+
+export LD_LIBRARY_PATH="\
+$NVIDIA_LIB_BASE/cublas/lib:\
+$NVIDIA_LIB_BASE/cudnn/lib:\
+$NVIDIA_LIB_BASE/cuda_runtime/lib:\
+$LD_LIBRARY_PATH"
+
+echo "CUDA library paths set."

requirements.txt

@@ -1,3 +1,7 @@
 numpy
 sounddevice
 faster-whisper
+transformers 
+torch
+sentencepiece
+sacremoses

test_live.py

@@ -11,7 +11,7 @@ CHUNK_SECONDS = 3        # record N seconds, then transcribe
 DEVICE = None            # None = default mic, or set to device index
 
 # --- Setup ---
-model = WhisperModel(MODEL_SIZE, device="cpu", compute_type="int8")
+model = WhisperModel(MODEL_SIZE, device="cuda", compute_type="int8")
 audio_queue = queue.Queue()
 
 def audio_callback(indata, frames, time, status):

test_live_translate.py

@@ -0,0 +1,49 @@
+import time
+import torch
+from transformers import MarianMTModel, MarianTokenizer
+
+MODEL_NAME = "Helsinki-NLP/opus-mt-en-fr"
+tokenizer = MarianTokenizer.from_pretrained(MODEL_NAME)
+model = MarianMTModel.from_pretrained(MODEL_NAME)
+
+DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
+model = model.to(DEVICE)
+model.eval()  # disable dropout, slightly faster inference
+
+FILE_PATH = "input.txt"
+OUTPUT_PATH = "output_fr.txt"
+
+def translate(text: str) -> str:
+    inputs = tokenizer(
+        text,
+        return_tensors="pt",
+        padding=True,
+        truncation=True,
+        max_length=512
+    ).to(DEVICE)
+
+    with torch.no_grad():  # saves memory, faster on GPU
+        translated = model.generate(**inputs)
+
+    return tokenizer.decode(translated[0], skip_special_tokens=True)
+
+def tail_and_translate(filepath: str):
+    with open(filepath, "r", encoding="utf-8") as f:
+        f.seek(0, 2)  # jump to end of file
+        print(f"Watching {filepath}...")
+
+        with open(OUTPUT_PATH, "a", encoding="utf-8") as out:
+            while True:
+                line = f.readline()
+                if line:
+                    line = line.strip()
+                    if line:
+                        translated = translate(line)
+                        print(f"EN: {line}")
+                        print(f"FR: {translated}")
+                        out.write(translated + "\n")
+                        out.flush()
+                else:
+                    time.sleep(0.2)
+
+tail_and_translate(FILE_PATH)