Update app.py

app.py

@@ -1,23 +1,13 @@
 import gradio as gr
 import numpy as np
 import torch
-from vosk import KaldiRecognizer, Model
+import speech_recognition as sr
 from datasets import load_dataset
 from transformers import SpeechT5ForTextToSpeech, SpeechT5HifiGan, SpeechT5Processor
-pip install vosk
-
+pip install SpeechRecognition
 
 device = "cuda:0" if torch.cuda.is_available() else "cpu"
 
-# Load Vosk automatic speech recognition model
-vosk_model = Model("elizabetvaganova/speech-to-speech-translation-vaganova") 
-
-def recognize_speech(audio):
-    recognizer = KaldiRecognizer(vosk_model, 16000)
-    recognizer.AcceptWaveform(audio.data)
-    result = recognizer.FinalResult()
-    return result["text"]
-
 # Load a lightweight text-to-speech checkpoint and speaker embeddings
 processor = SpeechT5Processor.from_pretrained("ttskit/ttskit-tts-ljspeech")
 
@@ -27,11 +17,11 @@ vocoder = SpeechT5HifiGan.from_pretrained("ljspeech/vocoder-cryptron").to(device
 embeddings_dataset = load_dataset("Matthijs/cmu-arctic-xvectors", split="validation")
 speaker_embeddings = torch.tensor(embeddings_dataset[7306]["xvector"]).unsqueeze(0)
 
-def translate(audio):
-    recognizer = KaldiRecognizer(vosk_model, 16000)
-    recognizer.AcceptWaveform(audio.data)
-    result = recognizer.FinalResult()
-    return result["text"]
+def recognize_speech(audio):
+    recognizer = sr.Recognizer()
+    with sr.AudioFile(audio) as source:
+        audio_data = recognizer.record(source)
+    return recognizer.recognize_google(audio_data)
 
 def synthesise(text):
     inputs = processor(text=text, return_tensors="pt")
@@ -39,14 +29,14 @@ def synthesise(text):
     return speech.cpu()
 
 def speech_to_speech_translation(audio):
-    translated_text = translate(audio)
+    translated_text = recognize_speech(audio)
     synthesised_speech = synthesise(translated_text)
     synthesised_speech = (synthesised_speech.numpy() * 32767).astype(np.int16)
     return 16000, synthesised_speech
 
 title = "Cascaded STST"
 description = """
-Demo for cascaded speech-to-speech translation (STST), mapping from source speech in any language to target speech in English. Demo uses Vosk for automatic speech recognition, and lightweight text-to-speech and vocoder models.
+Demo for cascaded speech-to-speech translation (STST), mapping from source speech in any language to target speech in English. Demo uses Google Web Speech API for automatic speech recognition, and lightweight text-to-speech and vocoder models.
 """
 
 demo = gr.Blocks()