diff --git a/transcribe.py b/transcribe.py
index f921f8b..c497c27 100644
--- a/transcribe.py
+++ b/transcribe.py
@@ -25,8 +25,9 @@ from silero_vad import load_silero_vad, get_speech_timestamps
 from transformers import MarianMTModel, MarianTokenizer
 
 # Parameters
-WHISPER_MODEL = "mlx-community/whisper-small.en-mlx"
-# List of language pairs (English to ...)
+# Using the multilingual small model
+WHISPER_MODEL = "mlx-community/whisper-small-mlx" 
+# Translation models (English -> Target)
 TARGET_LANGS = {
     "Spanish": "Helsinki-NLP/opus-mt-en-es",
     "French":  "Helsinki-NLP/opus-mt-en-fr",
@@ -51,8 +52,8 @@ def main():
     device = "mps" if torch.backends.mps.is_available() else "cpu"
     print(f"Using device: {device}")
 
-    # 1. Load models first
-    print(f"Loading Whisper model '{WHISPER_MODEL}'...")
+    # 1. Load models
+    print(f"Loading Multilingual Whisper model '{WHISPER_MODEL}'...")
     
     translation_engines = {}
     for lang_name, model_id in TARGET_LANGS.items():
@@ -76,13 +77,12 @@ def main():
         print("Invalid input, using default device.")
         device_index = None
 
-    print(f"\nStarting live transcription & multiple translations using device {device_index if device_index is not None else 'default'}... (Press Ctrl+C to stop)")
+    print(f"\nStarting Multilingual live transcription... (Press Ctrl+C to stop)")
     
     audio_buffer = []
     speech_started = False
 
     try:
-        # 3. Start stream
         with sd.InputStream(samplerate=SAMPLERATE, channels=CHANNELS, callback=callback, blocksize=BLOCK_SIZE, device=device_index):
             while True:
                 while not audio_queue.empty():
@@ -108,23 +108,35 @@ def main():
                         
                         if (buffer_len_samples - last_end) > (SAMPLERATE * MIN_SILENCE_DURATION_MS / 1000) or buffer_len_samples > BUFFER_LIMIT:
                             
-                            # Transcribe once
-                            result = mlx_whisper.transcribe(current_audio, path_or_hf_repo=WHISPER_MODEL)
-                            original_text = result['text'].strip()
+                            # 1. Transcribe & Detect Language
+                            # We use task="transcribe" to get the original text and detect language
+                            transcription_result = mlx_whisper.transcribe(current_audio, path_or_hf_repo=WHISPER_MODEL)
+                            original_text = transcription_result['text'].strip()
+                            detected_lang = transcription_result.get('language', 'unknown')
                             
                             if original_text:
-                                if len(original_text) > 250:
-                                    original_text = original_text[:247] + "..."
+                                print(f"\n[{detected_lang.upper()} detected]: {original_text}")
                                 
-                                print(f"\n[EN]: {original_text}")
-                                
-                                # Translate to all targets
-                                for lang_name, (model, tokenizer) in translation_engines.items():
-                                    inputs = tokenizer(original_text, return_tensors="pt", padding=True).to(device)
-                                    with torch.no_grad():
-                                        translated_tokens = model.generate(**inputs, max_new_tokens=150)
-                                    translated_text = tokenizer.decode(translated_tokens[0], skip_special_tokens=True)
-                                    print(f"[{lang_name[:2].upper()}]: {translated_text}")
+                                # 2. Bridge to English if not already English
+                                if detected_lang != "en":
+                                    # Use Whisper to translate the segment to English
+                                    bridge_result = mlx_whisper.transcribe(current_audio, path_or_hf_repo=WHISPER_MODEL, task="translate")
+                                    english_text = bridge_result['text'].strip()
+                                    print(f"[EN Bridge]: {english_text}")
+                                else:
+                                    english_text = original_text
+
+                                # 3. Translate from English to other languages
+                                if english_text:
+                                    if len(english_text) > 250:
+                                        english_text = english_text[:247] + "..."
+                                    
+                                    for lang_name, (model, tokenizer) in translation_engines.items():
+                                        inputs = tokenizer(english_text, return_tensors="pt", padding=True).to(device)
+                                        with torch.no_grad():
+                                            translated_tokens = model.generate(**inputs, max_new_tokens=150)
+                                        translated_text = tokenizer.decode(translated_tokens[0], skip_special_tokens=True)
+                                        print(f"[{lang_name[:2].upper()}]: {translated_text}")
                             
                             audio_buffer = []
                             speech_started = False