Update app.py

app.py

@@ -1,6 +1,8 @@
 import gradio as gr
 from transformers import pipeline, AutoTokenizer
 import numpy as np
+import torch
+import scipy
 
 # Load the pipeline for speech recognition and translation
 pipe = pipeline(
@@ -9,7 +11,8 @@ pipe = pipeline(
     tokenizer="Baghdad99/saad-speech-recognition-hausa-audio-to-text"
 )
 translator = pipeline("text2text-generation", model="Baghdad99/saad-hausa-text-to-english-text")
-tts = pipeline("text-to-speech", model="Baghdad99/english_voice_tts")
+model = VitsModel.from_pretrained("facebook/mms-tts-eng")
+tokenizer = AutoTokenizer.from_pretrained("facebook/mms-tts-eng")
 
 # Define the function to translate speech
 def translate_speech(audio):
@@ -39,19 +42,16 @@ def translate_speech(audio):
         print("The translated text does not contain 'generated_token_ids'")
         return
 
-    # 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
+    # Use the VITS model to synthesize the translated text into speech
+    inputs = tokenizer(translated_text_str, return_tensors="pt")
+    with torch.no_grad():
+        output = model(**inputs).waveform
 
-    # Check if the synthesised speech contains 'audio'
-    if 'audio' in synthesised_speech:
-        synthesised_speech_data = synthesised_speech['audio']
-    else:
-        print("The synthesised speech does not contain 'audio'")
-        return
+    # Save the synthesized speech to a WAV file
+    scipy.io.wavfile.write("synthesized_speech.wav", rate=model.config.sampling_rate, data=output.float().numpy())
 
     # Scale the audio data to the range of int16 format
-    synthesised_speech = (synthesised_speech_data * 32767).astype(np.float32)
+    synthesised_speech = (output * 32767).astype(np.int16)
 
     return 16000, synthesised_speech