Update app.py

app.py

@@ -10,7 +10,12 @@ pipe = pipeline(
     model="DrishtiSharma/whisper-large-v2-hausa",
     tokenizer="DrishtiSharma/whisper-large-v2-hausa"
 )
-translator = pipeline("text2text-generation", model="Baghdad99/saad-hausa-text-to-english-text")
+
+# Load the new translation model and tokenizer
+model_name = 'jbochi/madlad400-3b-mt'
+model = T5ForConditionalGeneration.from_pretrained(model_name)
+tokenizer = T5Tokenizer.from_pretrained(model_name)
+
 tts = pipeline("text-to-speech", model="Baghdad99/english_voice_tts")
 
 # Define the function to translate speech
@@ -51,9 +56,16 @@ def translate_speech(audio_file):
         print("The translated text does not contain 'generated_token_ids'")
         return
 
+   # Use the new translation model to translate the transcription
+    text = "translate Hausa to English: " + transcription
+    input_ids = tokenizer.encode(text, return_tensors="pt")
+    outputs = model.generate(input_ids=input_ids)
+
+    # Decode the tokens into text
+    translated_text_str = tokenizer.decode(outputs[0], skip_special_tokens=True)
+
     # Use the text-to-speech pipeline to synthesize the translated text
     synthesised_speech = tts(translated_text_str)
-    print(f"Synthesised speech: {synthesised_speech}")  # Print the synthesised speech to see what it contains
 
     # Check if the synthesised speech contains 'audio'
     if 'audio' in synthesised_speech:
@@ -71,7 +83,6 @@ def translate_speech(audio_file):
     return 16000, synthesised_speech
 
 
-
 # Define the Gradio interface
 iface = gr.Interface(
     fn=translate_speech,