Upload app.py

app.py

@@ -1,22 +1,49 @@
+# -*- coding: utf-8 -*-
+"""HW3_ml.ipynb
+
+Automatically generated by Colaboratory.
+
+Original file is located at
+    https://colab.research.google.com/drive/1z4ht7K9pttbgWmDDnrQhqoZ6SYAiaeUe
+"""
+
+# !pip -q uninstall gradio -y
+# !pip -q install gradio==3.50.2
+
+# !pip -q install datasets
+
 import gradio as gr
 import numpy as np
 import torch
 from datasets import load_dataset
 
-from transformers import SpeechT5ForTextToSpeech, SpeechT5HifiGan, SpeechT5Processor, pipeline, WhisperProcessor, VitsModel, VitsTokenizer
-
+from transformers import SpeechT5ForTextToSpeech, SpeechT5HifiGan, SpeechT5Processor, pipeline, WhisperProcessor
 
 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)
+asr_pipe = pipeline("automatic-speech-recognition", model="voidful/wav2vec2-xlsr-multilingual-56", device=device)
 
+# !pip -q install sentencepiece
 # load text-to-speech checkpoint and speaker embeddings
-processor = WhisperProcessor.from_pretrained("openai/whisper-small", language="russian")
+# processor = SpeechT5Processor.from_pretrained("microsoft/speecht5_tts")
+processor = WhisperProcessor.from_pretrained(
+    "openai/whisper-small")
+
+translator1 = pipeline("translation", model="Helsinki-NLP/opus-mt-mul-en")
+translator2 = pipeline("translation", model="Helsinki-NLP/opus-mt-en-ru")
+
+from transformers import VitsModel, VitsTokenizer
+
+# model = pipeline("text-to-speech", model="suno/bark-small")
 
 model = VitsModel.from_pretrained("facebook/mms-tts-rus")
 tokenizer = VitsTokenizer.from_pretrained("facebook/mms-tts-rus")
 
+def translator_mul_ru(text):
+
+    translation  = translator2(translator1(text)[0]['translation_text'])
+    return translation[0]['translation_text']
 
 def translate(audio):
     outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "translate"})
@@ -24,25 +51,27 @@ def translate(audio):
 
 
 def synthesise(text):
-    inputs = processor(text=text, return_tensors="pt")["input_ids"]
+    translated_text = translator_mul_ru(text)
+    inputs = tokenizer(translated_text, return_tensors="pt")
+    input_ids = inputs["input_ids"]
+
     with torch.no_grad():
-        outputs = model(inputs)
+        outputs = model(input_ids)
     speech = outputs["waveform"]
     return speech.cpu()
 
 
 def speech_to_speech_translation(audio):
     translated_text = translate(audio)
+    print(translated_text)
     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:
-
 ![Cascaded STST](https://huggingface.co/datasets/huggingface-course/audio-course-images/resolve/main/s2st_cascaded.png "Diagram of cascaded speech to speech translation")
 """
 
@@ -60,12 +89,12 @@ 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"]],
     title=title,
     description=description,
 )
 
 with demo:
-    gr.TabbedInterface([mic_translate, file_translate], ["Microphone", "Audio File"])
+    gr.TabbedInterface([mic_translate, file_translate], ["Microphone", "File"])
 
 demo.launch()
+