Update app.py

app.py

@@ -1,49 +1,43 @@
 import gradio as gr
 import numpy as np
 import torch
-from datasets import load_dataset
-
-from transformers import SpeechT5ForTextToSpeech, SpeechT5HifiGan, SpeechT5Processor, pipeline
 
+from transformers import AutoTokenizer, VitsModel
+from transformers import pipeline
 
 device = "cuda:0" if torch.cuda.is_available() else "cpu"
 
-# load speech translation checkpoint
-asr_pipe = pipeline("automatic-speech-recognition", model="openai/whisper-base", device=device)
-
-# load text-to-speech checkpoint and speaker embeddings
-processor = SpeechT5Processor.from_pretrained("microsoft/speecht5_tts")
-
-model = SpeechT5ForTextToSpeech.from_pretrained("microsoft/speecht5_tts").to(device)
-vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan").to(device)
+# Translate audio to russian text
+asr_pipe = pipeline("automatic-speech-recognition", model="openai/whisper-tiny", device=device)
+translator_to_ru = pipeline("translation", model="Helsinki-NLP/opus-mt-en-ru")
 
-embeddings_dataset = load_dataset("Matthijs/cmu-arctic-xvectors", split="validation")
-speaker_embeddings = torch.tensor(embeddings_dataset[7306]["xvector"]).unsqueeze(0)
-
-
-def translate(audio):
+def translate(audio, translator_to_ru: pipeline = translator_to_ru):
     outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "translate"})
-    return outputs["text"]
+    return translator_to_ru(outputs['text'])[0]['translation_text']
 
+# Text to russian speech
+model = VitsModel.from_pretrained("facebook/mms-tts-rus")
+tokenizer = AutoTokenizer.from_pretrained("facebook/mms-tts-rus")
 
-def synthesise(text):
-    inputs = processor(text=text, return_tensors="pt")
-    speech = model.generate_speech(inputs["input_ids"].to(device), speaker_embeddings.to(device), vocoder=vocoder)
-    return speech.cpu()
+def synthesise(text: str, tokenizer: AutoTokenizer = tokenizer, model: VitsModel = model):
+    inputs = tokenizer(text, return_tensors="pt")
+    # print(inputs)
+    with torch.no_grad():
+        output = model(**inputs).waveform
+    return output.cpu()
 
 
 def speech_to_speech_translation(audio):
     translated_text = translate(audio)
     synthesised_speech = synthesise(translated_text)
     synthesised_speech = (synthesised_speech.numpy() * 32767).astype(np.int16)
-    return 16000, synthesised_speech
+    return 16000, synthesised_speech[0]
 
 
 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 OpenAI's [Whisper Base](https://huggingface.co/openai/whisper-base) model for speech translation, and Microsoft's
-[SpeechT5 TTS](https://huggingface.co/microsoft/speecht5_tts) model for text-to-speech:
-
+Demo for cascaded speech-to-speech translation (STST), mapping from source speech in multi language to target speech in Russian. Demo uses OpenAI's [Whisper Tiny](https://huggingface.co/openai/whisper-tiny) model for speech translation, and Facebook's
+[mms-tts-rus](https://huggingface.co/acebook/mms-tts-rus) model for text-to-speech:
 ![Cascaded STST](https://huggingface.co/datasets/huggingface-course/audio-course-images/resolve/main/s2st_cascaded.png "Diagram of cascaded speech to speech translation")
 """
 
@@ -61,7 +55,7 @@ file_translate = gr.Interface(
     fn=speech_to_speech_translation,
     inputs=gr.Audio(source="upload", type="filepath"),
     outputs=gr.Audio(label="Generated Speech", type="numpy"),
-    examples=[["./example.wav"]],
+    examples=[["./test_2.wav"]],
     title=title,
     description=description,
 )
@@ -69,4 +63,4 @@ file_translate = gr.Interface(
 with demo:
     gr.TabbedInterface([mic_translate, file_translate], ["Microphone", "Audio File"])
 
-demo.launch()
+demo.launch()