Added Italian to French Feature (#1)

- Added Italian to French Feature (b08b273c06fcf98751168eea77a01fecffd2ed22)

app.py

@@ -3,7 +3,12 @@ import numpy as np
 import torch
 from datasets import load_dataset
 
+
+from transformers import AutoTokenizer, TFMarianMTModel
+from typing import List
+
 from transformers import SpeechT5ForTextToSpeech, SpeechT5HifiGan, SpeechT5Processor, pipeline
+from transformers import SpeechT5Processor, SpeechT5ForTextToSpeech, SpeechT5HifiGan, AutoProcessor
 
 
 device = "cuda:0" if torch.cuda.is_available() else "cpu"
@@ -12,18 +17,34 @@ device = "cuda:0" if torch.cuda.is_available() else "cpu"
 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")
+processor = AutoProcessor.from_pretrained("Sandiago21/speecht5_finetuned_facebook_voxpopuli_french")
 
-model = SpeechT5ForTextToSpeech.from_pretrained("microsoft/speecht5_tts").to(device)
+model = SpeechT5ForTextToSpeech.from_pretrained("Sandiago21/speecht5_finetuned_facebook_voxpopuli_french").to(device)
 vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan").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):
     outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "translate"})
-    return outputs["text"]
+    english =  outputs["text"]
+
+    src = "en"  # source language
+    trg = "fr"  # target language
+    model_name = f"Helsinki-NLP/opus-mt-{src}-{trg}"
+
+    model = TFMarianMTModel.from_pretrained(model_name)
+    tokenizer = AutoTokenizer.from_pretrained(model_name)
+
+    batch = tokenizer([english], return_tensors="tf")
+    gen = model.generate(**batch)
+    return tokenizer.batch_decode(gen, skip_special_tokens=True)[0]
+
+
 
 
 def synthesise(text):
@@ -43,7 +64,6 @@ 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")
 """
 
@@ -61,7 +81,7 @@ 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=[["./example.wav"]],
     title=title,
     description=description,
 )
@@ -69,4 +89,4 @@ file_translate = gr.Interface(
 with demo:
     gr.TabbedInterface([mic_translate, file_translate], ["Microphone", "Audio File"])
 
-demo.launch()
+demo.launch()