Update app.py

app.py

@@ -1,7 +1,7 @@
 import torch
 import librosa
 import gradio as gr
-from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor, pipeline, AutoTokenizer
+from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor, AutoProcessor, SeamlessM4Tv2Model
 import numpy as np
 import soundfile as sf
 import tempfile
@@ -9,7 +9,11 @@ import tempfile
 # Load the models and processors
 asr_model = Wav2Vec2ForCTC.from_pretrained("jonatasgrosman/wav2vec2-large-xlsr-53-english")
 asr_processor = Wav2Vec2Processor.from_pretrained("jonatasgrosman/wav2vec2-large-xlsr-53-english")
-translator = pipeline("text2text-generation", model="dammyogt/damilola-finetuned-NLP-opus-mt-en-ha")
+
+# Load the SeamlessM4T model and processor
+translator_model = SeamlessM4Tv2Model.from_pretrained("facebook/seamless-m4t-v2-large")
+translator_processor = AutoProcessor.from_pretrained("facebook/seamless-m4t-v2-large")
+
 tts = pipeline("text-to-speech", model="Baghdad99/hausa_voice_tts")
 
 def translate_speech(audio_file_path):
@@ -29,22 +33,19 @@ def translate_speech(audio_file_path):
     transcription = asr_processor.decode(pred_ids)
     print(f"Transcription: {transcription}")  # Print the transcription
 
-    # Use the translation pipeline to translate the transcription
-    translated_text = translator(transcription, return_tensors="pt")
-    print(f"Translated text: {translated_text}")  # Print the translated text
+    # Prepare the input dictionary for the translator
+    text_inputs = translator_processor(text=transcription, src_lang="eng", return_tensors="pt")
 
-    # Check if the translated text contains 'generated_token_ids'
-    if 'generated_token_ids' in translated_text[0]:
-        # Decode the tokens into text
-        translated_text_str = translator.tokenizer.decode(translated_text[0]['generated_token_ids'])
-        
-        # Remove special tokens
-        translated_text_str = translated_text_str.replace("<pad>", "").replace("</s>", "").strip()
+    # Use the translator model to translate the transcription
+    translated_text = translator_model.generate(**text_inputs, tgt_lang="ha")  # Change the target language to Hausa
 
-        print(f"Translated text string: {translated_text_str}")  # Print the translated text string
-    else:
-        print("The translated text does not contain 'generated_token_ids'")
-        return
+    # Decode the translated text
+    translated_text_str = translator_processor.decode(translated_text[0])
+
+    # Remove special tokens
+    translated_text_str = translated_text_str.replace("<pad>", "").replace("</s>", "").strip()
+
+    print(f"Translated text string: {translated_text_str}")  # Print the translated text string
 
     # Use the text-to-speech pipeline to synthesize the translated text
     synthesised_speech = tts(translated_text_str)
@@ -63,14 +64,3 @@ def translate_speech(audio_file_path):
     synthesised_speech = (synthesised_speech_data * 32767).astype(np.int16)
 
     return 16000, synthesised_speech
-
-# Define the Gradio interface
-iface = gr.Interface(
-    fn=translate_speech, 
-    inputs=gr.inputs.Audio(type="filepath"),  # Change this line
-    outputs=gr.outputs.Audio(type="numpy"),
-    title="English to Hausa Translation",
-    description="Realtime demo for English to Hausa translation using speech recognition and text-to-speech synthesis."
-)
-
-iface.launch()