Update app.py

app.py

@@ -1,44 +1,54 @@
+# -*- coding: utf-8 -*-
+"""app.ipynb
+Automatically generated by Colaboratory.
+Original file is located at
+    https://colab.research.google.com/drive/16MxXQeF3O0htL9eQ61aa6ZxnApGg9TKN
+"""
+
 import gradio as gr
 import numpy as np
 import torch
 
-from transformers import AutoTokenizer, VitsModel
-from transformers import pipeline
+from transformers import pipeline, VitsModel, VitsTokenizer, FSMTForConditionalGeneration, FSMTTokenizer
 
 device = "cuda:0" if torch.cuda.is_available() else "cpu"
 
-# 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")
+#eng audio to text transformation
+asr_pipe = pipeline("automatic-speech-recognition", model="asapp/sew-d-tiny-100k-ft-ls100h", device=device)
+
+#eng text to rus text translation
+translation_pipe = pipeline("translation", model="Helsinki-NLP/opus-mt-en-ru")
 
-def translate(audio, translator_to_ru: pipeline = translator_to_ru):
+#rus text to rus speech transformation
+vits_model = VitsModel.from_pretrained("facebook/mms-tts-rus")
+vits_tokenizer = VitsTokenizer.from_pretrained("facebook/mms-tts-rus")
+
+def transform_audio_to_speech_en(audio):
     outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "translate"})
-    return translator_to_ru(outputs['text'])[0]['translation_text']
+    return outputs["text"]
 
-# Text to russian speech
-model = VitsModel.from_pretrained("facebook/mms-tts-rus")
-tokenizer = AutoTokenizer.from_pretrained("facebook/mms-tts-rus")
+def translator(text):
+  translated_text = translation_pipe(text)
+  return translated_text[0]['translation_text']
 
-def synthesise(text: str, tokenizer: AutoTokenizer = tokenizer, model: VitsModel = model):
-    inputs = tokenizer(text, return_tensors="pt")
-    # print(inputs)
+def synthesise(translated_text):
+    translated_text = translator(translated_text)
+    inputs = vits_tokenizer(translated_text, return_tensors="pt")
     with torch.no_grad():
-        output = model(**inputs).waveform
-    return output.cpu()
-
+        speech = vits_model(**inputs).waveform
+    return speech.cpu()
 
 def speech_to_speech_translation(audio):
-    translated_text = translate(audio)
+    translated_text = transform_audio_to_speech_en(audio)
     synthesised_speech = synthesise(translated_text)
     synthesised_speech = (synthesised_speech.numpy() * 32767).astype(np.int16)
     return 16000, synthesised_speech[0]
 
-
 title = "Cascaded STST"
 description = """
-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")
+В Демо используется модель SEW-D-tiny(https://huggingface.co/asapp/sew-d-tiny-100k-ft-ls100h),
+распознающая английскую речь и преобразующая ее в строку. Затем с помощью модели Helsinki-NLP/opus-mt-en-ru(https://huggingface.co/Helsinki-NLP/opus-mt-en-ru) текст
+переводится на русский язык и преобразуется в русскую речь с помощью модели facebook/mms-tts-rus(https://huggingface.co/facebook/mms-tts-rus).
 """
 
 demo = gr.Blocks()
@@ -55,7 +65,7 @@ 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=[["./test_2.wav"]],
+    examples=[["./example.wav"]],
     title=title,
     description=description,
 )